Dispersion-Corrected Mean-Field Electronic Structure Methods

Apr 14, 2016 - His research concentrates on the development and application of fast electronic structure methods, with particular focus on organic cry...
15 downloads 11 Views 5MB Size
This is an open access article published under an ACS AuthorChoice License, which permits copying and redistribution of the article or any adaptations for non-commercial purposes.

Review pubs.acs.org/CR

Dispersion-Corrected Mean-Field Electronic Structure Methods Stefan Grimme,* Andreas Hansen, Jan Gerit Brandenburg, and Christoph Bannwarth Mulliken Center for Theoretical Chemistry, Universität Bonn, 53113 Bonn, Germany ABSTRACT: Mean-field electronic structure methods like Hartree−Fock, semilocal density functional approximations, or semiempirical molecular orbital (MO) theories do not account for long-range electron correlation (London dispersion interaction). Inclusion of these effects is mandatory for realistic calculations on large or condensed chemical systems and for various intramolecular phenomena (thermochemistry). This Review describes the recent developments (including some historical aspects) of dispersion corrections with an emphasis on methods that can be employed routinely with reasonable accuracy in large-scale applications. The most prominent correction schemes are classified into three groups: (i) nonlocal, density-based functionals, (ii) semiclassical C6-based, and (iii) one-electron effective potentials. The properties as well as pros and cons of these methods are critically discussed, and typical examples and benchmarks on molecular complexes and crystals are provided. Although there are some areas for further improvement (robustness, many-body and short-range effects), the situation regarding the overall accuracy is clear. Various approaches yield long-range dispersion energies with a typical relative error of 5%. For many chemical problems, this accuracy is higher compared to that of the underlying mean-field method (i.e., a typical semilocal (hybrid) functional like B3LYP).

CONTENTS 1. Introduction 2. Setting the Stage 3. Older Dispersion Corrections 3.1. Dispersion-corrected DFT 3.2. Dispersion-corrected HF 4. Modern Dispersion Corrections 4.1. Semiclassical treatments of the dispersion interaction 4.1.1. D2 approach 4.1.2. D3 approach 4.1.3. Tkatchenko−Scheffler model 4.1.4. TS-based many-body dispersion scheme 4.1.5. Dispersion corrections based on maximally localized Wannier functions 4.1.6. Exchange-dipole moment model 4.1.7. Density-dependent energy correction 4.1.8. Local-response dispersion model 4.2. Nonlocal density-based dispersion corrections 4.2.1. van der Waals density functionals 4.2.2. Vydrov and Van Voorhis functionals 4.3. Effective one-electron potentials and further aspects 4.3.1. External potentials 4.3.2. Semilocal functionals 4.3.3. Dispersion-mimicking and dispersioncompensating effects 4.3.4. Dispersion in periodic systems 4.3.5. Dispersion-corrected semiempirical MO methods 4.4. Methodological perspective 5. Typical Applications and Benchmarks © 2016 American Chemical Society

5.1. Exemplifying the distance regimes of the dispersion energy 5.2. Peptide conformation potential 5.3. Benzene crystal 5.4. Standard benchmark sets for noncovalent interactions 6. Summary and Outlook Author Information Corresponding Author Notes Biographies Acknowledgments References

5105 5108 5111 5111 5112 5113 5113 5115 5116 5118 5119 5120 5121 5122 5122

5135 5137 5139 5139 5143 5144 5144 5144 5144 5144 5144

1. INTRODUCTION Mean-field (MF) electronic structure methods like Hartree− Fock (HF), approximate, semilocal Kohn−Sham density functional theory (KS-DFT), and semiempirical molecular orbital (SE-MO) theory are widely used in chemistry and physics, particularly for large systems. These methods do not describe long-range electronic correlation effects, and hence they cannot account for so-called London dispersion interactions.1−5 The development of appropriate dispersion corrections to cure this deficiency is a very active and practically relevant field of theoretical research. The increased attention to this topic in the past few years is directly reflected in the growing number of citations found in the field of dispersion correction or van der Waals + density f unctional theory (see Figure 1). This Review shall report

5123 5123 5125 5127 5127 5128 5129 5131

Special Issue: Noncovalent Interactions

5131 5132 5135

Received: September 11, 2015 Published: April 14, 2016 5105

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

It has now become very clear, especially for the chemistry and physics of large or condensed-phase systems, e.g., in bio- or nanoarchitectures, that inclusion of these interactions in theoretical simulations is indispensable in order to reach chemical accuracy (∼1 kcal/mol). Classical, atomistic forcefields have treated these interactions from the very beginning, for example, by interatomic Lennard-Jones type 6−12 potentials. For a recent overview on various calculation methods for large systems, see ref 18. Because dispersion effects are due to ubiquitous electron correlation, they also influence the accuracy of theoretical (reaction) thermodynamics.19 See ref 20 for a recent review on dispersion effects from a chemical/catalysis point of view. They are, of course, most important when aiming at a good description of intermolecular, noncovalent interactions (NCIs), and this is also historically the origin of the field. However, here we will pursue a more general picture of dispersion effects as outlined in the next section and illustrated with examples in section 5. If perturbation theory is applied to the electrons of interacting atoms,4,21 two distance regimes can be identified (Figure 2). At

Figure 1. Number of citations found on the Web of Science6 in the time period from 2000 through 2015 for papers on “dispersion correction” or “density functional theory” + “van der Waals”.

on dispersion corrections for MF approaches that have been developed and applied in the last few years. For related works with some review character, see refs 7−16. Dispersion interactions can be empirically defined as the attractive part of a van der Waals (vdW)-type interaction potential between atoms or molecules that are not directly (covalent or ionic) bonded to each other. They are prominently mentioned in the very first quantum chemistry textbook by Hellmann.17 In the current literature, the terms “dispersion” and “vdW” are often used synonymously. The attribute dispersive should be avoided in this context because it has various other meanings in science. According to a more precise definition, London dispersion interactions result from relatively longranged electron correlation effects in any many-electron system that neither requires “polarity” nor wave function (WF) overlap (see section 5 for a discussion on the distance regime relevant for London dispersion). They involve coupled local components and can be explained in a stationary, time-independent electronic state picture. Already in 1930,1 from the simplest perturbation theory Eisenschitz and London have derived the famous asymptotic formula for the dispersion energy between two atoms A and B at large distance R 3 IAIB 0 0 −6 AB AB Edisp ≈− αAαB R = −C6,approx R−6 2 IA + IB (1)

Figure 2. Qualitative behavior of the correlation energy at short distances R (unified atom limit) and asymptotically for large distances, where it is termed interatomic (molecular) London dispersion energy (k1, k2, and C6 are system-specific constants). The dashed line depicts the interpolation that is used in many correction methods and indicates the medium-range correlation/dispersion range of ∼2.5−3.5 Å. Reprinted with permission from ref 14. Copyright 2014 Wiley-VCH.

large distances, the dispersion energy is given by the well-known −1/R6 dependence on the interaction distance. Note that the C6 dispersion coefficient can refer to atoms as well as entire systems (molecules) because it is derived from a centered multipole expansion. In this long-distance region, London dispersion can be considered as the attractive part of a typical Lennard-Jones (model) potential. According to a less well-known analysis,21 at small distances, the dispersion energy becomes constant and part of the normal correlation energy (as first used in a correction scheme by Becke and Johnson22). The currently established notion is that the dispersion energy is a continuous quantity representing a meaningful concept for all values of R (i.e., including the dashed line in Figure 2). It can be expected that in intermediate regions it influences the electronic energy of molecules significantly and hence has to be considered in computational thermochemistry. Note that it is defined not only for the intermolecular/interatomic situation but also for the interaction of intramolecular fragments (functional groups)

where α0 is the static dipole polarizability and I is the atomic ionization potential. The atomic constants can be condensed to the pair-specific C6 dispersion coefficient, which determines the strength of the interaction. Note that Edisp is attractive for any distance and hence stabilizes molecules with respect to their constituting atoms, condensed phases over the (diluted) gas phase, and in general more dense structures and materials. The well-established London formula represents a central element in the correction schemes discussed herein. Quantum chemical methods to compute the C6 and higher-order dispersion coefficients are briefly discussed in section 4. Note that there is no consensus in the literature if the sign in the above equation should be stated explicitly (and hence the C6 would be positive) or if it is included in the coefficients. In any case, the long-range dispersion energy as an electron correlation effect is always an energy-lowering (negative) quantity. In this Review, we adopt the former variant, i.e., with C6 being a positive quantity. 5106

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

years;23,36−39 however, their slow convergence to the CBS limit still persists. The correlated WF method with lowest scaling (formally 6(N 5)) is second-order Møller−Plesset perturbation theory (MP2).40 A modified variant (scaled opposite-spin MP2)41 exists that scales with 6(N 4). Even at the CBS limit, however, MP2 performs badly for dispersion-dominated interactions,42−44 in particular for chemically unsaturated systems. The reason for this is that the dispersion energy contribution to the supermolecular MP2 energy lacks intramolecular correlation effects and therefore describes the longrange correlation energy on the so-called uncoupled level (see below). For association reactions of two separate species (eq 9), this situation could be remedied by replacing the MP2 dispersion contribution with coupling-inclusive terms from time-dependent HF42 or time-dependent DFT (the latter yielding the MP2C approach45). These methods unfortunately require the definition of separate fragments and are consequently not generally applicable. To arrive at a generally applicable MP2-based approach, range separation of the correlation energy into short- and long-range parts is possible (similar to Figure 2).46 Another workable approach is to introduce different scalings of the same-spin and opposite-spin components in MP247 including further developed variants thereof.48 Recently, MP2 correlation at short range has been combined with a dispersion correction originally developed for MF approaches, which then describes the correlation at long range.49,50 Such methods are beyond the focus of this Review. However, one should be aware of the fact that, for an accurate description of dispersion interactions, cost-efficient correlated WF methods require modifications at long range as well, notwithstanding the mentioned slower convergence to the CBS limit. Consequently, in the foreseeable future, MF approaches (in particular sophisticated DFT methods) will be the only reasonable choice to routinely conduct gradient calculations on large systems with >100 atoms. According to Figure 3, the corrections considered herein can be classified into three groups: nonlocal density-based (which mostly include a correction to the electronic potential V), semiclassical (C6-based) schemes (which often apply corrections only to the total energy E), and effective one-electron potentials. Almost all of the current dispersion corrections include empirical elements in various ways. Hence, solid benchmarking on reliable experimental or high-level theoretical reference data is mandatory. For the inter- and intramolecular interactions and mostly main-group thermochemistry considered here, the current gold standard in quantum chemistry is the WF-based singles and doubles coupled-cluster method with perturbative triples (CCSD(T)), which yields accurate results of benchmark quality ( B3LYP > PBE > CCSD(T). Dispersion corrections should properly correct this behavior not only in the equilibrium and large-distance regime (here >3.5 Å) but also for short interatomic

of noncovalent interactions. Normally, MF methods employ a supermolecular computation of the interaction energy ΔEAB, ΔEAB = E(AB) − E(A) − E(B)

(9)

(7)

where E(AB) and E(A)/E(B) refer to the complex and noninteracting fragment total energies, respectively. This approach is rather general but has the disadvantage that the result can be contaminated by the basis set superposition error (BSSE) and that no insight into the nature of the interaction is obtained.61 On the other hand, SAPT applies PT starting with unperturbed monomer WFs (HF or KS-DFT) and a doubleperturbation scheme in the monomer correlation and the intermolecular interaction V̂ , (8) Ĥ = F ̂ + Ŵ + V̂ ̂ ̂ where F is the (KS-)Fock operator and W is the MP perturbation in the case of HF. Note that the standard Rayleigh−Schrödinger PT cannot be applied at short distances because of non-negligible exchange interactions, i.e., the (antisymmetrized) fragment product states are not eigenstates of the zeroth order Hamiltonian. These difficulties are well-documented in the

Figure 5. Computed (aug-cc-pV5Z AO basis) potential energy curves for (a) two interacting Ar atoms and (b) an Ar atom interacting with a positive point charge. 5109

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

distances that occur in “real” systems, for example, in strained molecular systems or under high-pressure conditions. In fact, HF and B3LYP yield a completely unbound Ar2 complex contrary to what is obtained at the reference level (Re = 3.78 Å with a De of 0.28 kcal/mol). Part b of the figure shows the induction energies of the same methods when an Ar atom is perturbed by a point charge. As a stabilizing, second-order property, the induction energy Eind is also considerably influenced by the monomer electronic excitation energies. These are generally lower for DF than for HF,77 which explains the observed order for Eind: HF > CCSD(T) > B3LYP > PBE. Note the overestimated polarizability for both DFs, which in fact perform worse than HF. However, the too low induction energy with approximate DFs partially cancels the too repulsive EEXR part so that, at least for polar systems, relatively good NCIs can result (if dispersion is properly accounted for by additional corrections). This is in fact a fundamental reason for the relatively accurate NCI energies computed by dispersion-corrected DFT, which is documented in more detail in section 5. The above picture is consistent with the results of a recent study of three-body (nonadditive) effects on NCIs in typical organic complexes.78 DFT-SAPT is an ideal tool for distinguishing the different types of interactions contributing to the noncovalent binding. Another related approach that allows for a similar decomposition is the effective fragment potential (EFP) method.79 In Figure 6,

molecular trimer ABC does not have the interaction energy as given by the sum of the three dimer energies AB, AC, and BC. We give a summary of this topic following the clear and comprehensive description by Dobson.80 In his work he introduced three types of dispersion nonadditivity, defined in general as the departure of the dispersion interaction from a sum of gas-phase based Cn terms between pairs of previously selected “centers” (usually atoms). These categories are termed type-A, type-B, and type-C nonadditivity, respectively. The type-A case refers to the change of the pairwise dispersion coefficient for the free atoms CAB n when they are bound in a molecule to other atoms. This change (typically a decrease) in the corresponding “atom-in-molecule” dispersion coefficient is caused by bonding (increase in atomic excitation energies); crowding of orbitals, which effectively causes a reduction of atomic volume, and, consequently, the atomic polarizability (see ref 14 for a numerical overview for almost all atoms in the periodic table). This type-A category was ignored in all early dispersion-correction schemes (see section 3) and was often dealt with semiempirically, for example, by choosing the optimal coefficients through minimization of the error of the pairwise calculation relative to accurate molecular-binding energies of a dispersion-bonded training set.81,82 All modern pairwise corrections like the XDM model,22,83,84 the D3 scheme,85 or Tkatchenko and Scheffler’s (TS) approach86 include type-A effects by making the coefficients dependent on atom-inmolecule multipole moments (XDM), geometric coordination number (D3), or an atomic volume (TS), respectively. In contrast to the type-A case, type-B effects are nonadditive in a strict sense. They occur because an additional polarizable center C can effectively screen the Coulomb interaction between a given pair of centers AB, thus altering their pairwise correlation energy and the total dispersion energy of the trimeric system. The lowest-order triple-dipole term leads, in the isotropic case, to a three-center angularly and RAB−3RAC−3RBC−3 dependent interaction energy often dubbed the Axilrod−Teller−Muto interaction (see section 4.1 for more details). In a diagrammatic representation, an infinite number of further terms (ring diagrams) like this arise from multiple response function insertions into all possible Coulomb interaction lines, leading to N-center contributions. Note that the terminology in the literature regarding nonadditivity is not fully consistent. Dobson defines the type-B nonpairwise interaction as present between more than two atomic centers, and this definition is adopted in the many-body-dispersion (MBD) scheme of Tkatchenko and co-workers.87 It could, however, also refer to the number of Coulomb-lines, i.e., correlating electrons. In WF theory description, type-B effects are not captured by pairwise theories like MP2 or SAPT2 but require at least coupling of first-order perturbed WF as in MP3, or to infinite order as in RPA or CCSD. Nonlocal density-dependent dispersion corrections like vdWDF288 or VV1089 described in section 4.2 also do not account for type-B effects. Semilocal density functional approximations (DFAs) provide short-range many-body XC effects, but their magnitude and even their sign strongly depend on the particular functional. The general agreement with CCSD(T) reference data for rare gas (RG) triatomics is found to be insufficient.90,91 Atom or fragment-based approaches assume an inherently localized electronic structure, i.e., that electrons can be ascribed to centers of fragments. On the contrary, type-C nonadditivity is an intrinsically quantum phenomenon that occurs in cases of (near) degeneracy and strong electronic delocalization. This causes zero-energy denominators in perturbation theory and

Figure 6. Contributions to the binding of a π-stacked uracil dimer calculated with DFT-SAPT via the asymptotically corrected AC-PBE0 functional. Here, the PBE0 potential is shifted to reproduce the correct ionization potential of the uracil monomer. The first-order contributions are the exchange repulsion EEXR and the electrostatic Ees; the secondorder induction Eind and dispersion Edisp include the mixed exchange type contributions as well.

we exemplify this decomposition for DFT-SAPT on the πstacked uracil dimer for which the corresponding first-order exchange repulsion EEXR and electrostatic Ees and the secondorder induction Eind and dispersion Edisp interactions are shown. As usual, the exchange repulsion is the largest (and often the only) repulsive contribution. The electrostatic and dispersion terms are of similar magnitude. Apparently, the London dispersion is a very important interaction in this system, and its accurate description is mandatory for accurate quantum chemical predictions. The dispersion energy (like the other NCI energy components in eq 9) has many-body nonadditive components that are not covered by the purely atomic or molecular pairwise treatments as in eq 1. This means, in the simplest example, that an atomic or 5110

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

favors large electronic response and electron density fluctuations. Such situations occur in metals, and perhaps the most striking consequence of type-C effects is the quite different and slower spatial decay of the asymptotic interaction energy between gapless objects separated by a distance D (e.g., E ∝ D−3 for parallel graphene sheets92), which cannot be obtained by any atom pairwise scheme. The nonlinear size-dependence of the interaction energy of differently sized fullerenes, which have a highly delocalized electronic structure, is also a consequence of type-C dispersion effects. Note, however, that they are only really significant for conjugated system of substantial size (e.g., a few hundred carbon atoms) and that, for example, C60 containing vdW complexes are relatively well (but not very accurately)93 described by pairwise approaches.95 The nonadditivity of vdW interactions in nanostructures has been analyzed by models that recover the exact zero- and high-frequency limits. The resulting size dependencies of the nonadditive contributions can both increase and decrease the interaction depending on the system.96,97 In Figure 7, we sketch the long-range perturbation theory of intermolecular interactions as used in all modern London

3. OLDER DISPERSION CORRECTIONS The fact that standard local density approximations (LDAs) and generalized gradient approximations (GGAs) yield an inconsistent or even unbound description of small vdW complexes was discovered by three groups in the mid-1990s.98−100 RG dimers were suggested even earlier as difficult test cases for density functional approximations,101 although in this work the DFT failure was not that clearly documented. In 1996, Meijer and Sprik102 presented the first clear analysis of the strong density functional dependence of the problem (i.e., LDA vs GGA functionals) for the “canonical” case of the benzene dimer and, furthermore, made the connection to errors in computed solidphase properties like mass density or lattice energy. General claims that semilocal functionals cannot describe long-range correlation forces were occasionally made,102,103 but without clear reference or theoretical explanation. Even in 2002 the situation was not clear as indicated by the summary of van Mourik and Gdanitz104 of DFT studies on RG dimers, which identified over-repulsive as well as overbinding approximations that were further found to be strongly dependent on the actual system. Those days also have been dominated by the simple and actually incomplete picture that dispersion forces are only relevant for the intermolecular situation, i.e., vdW complexes and condensed phases. The modern notion that intramolecular dispersion effects are especially important in large systems19 was undeveloped, at least in the theoretical chemistry community. Note, however, that such intramolecular effects were occasionally mentioned in the experimental literature; see refs 105 and 106 for very early notes. Dispersion corrections to DFT and HF were scarcely applied prior to the year 2000, and if so, mostly in various nonstandardized ways. These precursors to more modern approaches are reviewed in this section. A common feature of the methods described herein is that they do not target a general solution of the problem but present ad hoc solutions for a few systems at most, or sometimes even for one specific vdW complex only. All methods reviewed here approximate the total molecular energy as a sum of a quantum mechanical (QM) energy from a mean-field approach EMF and a dispersion-energy contribution Edisp.

Figure 7. Three main approximations used in the long-range perturbation theory of interfragment interactions. This is the foundation of all modern London dispersion corrections, which treat the expansions to various orders. Here, the blue ν represents the noncategorized perturbation, while the green ν′ and the red ν″ correspond to induction (excitations only within one fragment) and dispersion (excitations within both fragments) interactions, respectively. ρA is the electron density on fragment A, qA is the monopole, μA the dipole, and θA is the quadrupole term arising from the multipole expansion of the Coulomb potential due to the electrons on A.

Etot = EMF + Edisp

(10)

where MF (i.e., mean-field approach) stands for Hartree−Fock (HF), a density functional approximation (DFA), or a SE-MO method. The EMF can be obtained even in a very approximate, nonself-consistent way. The dispersion coefficients (see below) in these early methods were not varied (geometry or electronic structure dependent) or computed specifically but were mostly taken as fixed parameters with numerical values often obtained in experiments. In the terminology of Dobson (see previous section), type-A (as well as type-B and type-C) nonadditivity effects were neglected.

dispersion approximations. For a system of interacting fragments (A, B, C, ...) with nonoverlapping electron densities, these approximations are (1) the interaction is decomposed into additive multibody interactions, (2) the interaction is treated perturbatively (London dispersion terms arising in second and higher order), and (3) the perturbing Coulomb potential is approximated by its multipole components. Due to the orthogonality of the excited states with respect to the ground state (the integral of the charge transition density is zero), the leading order corresponds to a dipole−dipole second-order perturbation of the two-fragment interaction.

3.1. Dispersion-corrected DFT

Cohen and Pack were likely the first who applied an atom pairwise dispersion correction to a density functional type interatomic potential.107 The authors used the so-called Gordon−Kim (GK)108 model as an approximation to HF and applied a dispersion correction of the form 5111

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews Edisp(R ) = −

C6

Review



6

R

C8 8

R



⎧ E DFT(R ) R < Rλ ⎪ Etot = ⎨ ⎪ ⎩ E HF(R ) + Ecorr,tot(R ) R > Rλ

C10 10

(11)

R

where R denotes an interatomic distance. The GK method uses the simple sum of atomic electron densities ρA and ρB to express the total molecular density ρ = ρA + ρB for a diatomic AB and employs various density functionals to compute electronic Coulomb, exchange, and correlation energies nonself-consistently. For the pair-dispersion coefficients Cn(A,B), general combination rules of the form C4 + 2l(A , B) =



fml (A)fnl (B)

n,m

ϵm(A)ϵn(B)(ϵm(A) + ϵn(B))



The resulting two-dimensional potentials compared well with the best known data from SAPT calculations, and derived thermal transport properties of RG/CO mixtures were in good agreement with corresponding measurements. Although the method was quite successful, it has never been applied to more complicated systems. Presumably, this is because the complexity of the problem when aiming at very high accuracy is hidden in the angle-dependent dispersion coefficients Cn(θ), which were taken from literature data but not actually computed specifically for the considered system. The common, textbook-prevalent view that dispersion forces are relatively unimportant for polar or hydrogen-bonded systems (where electrostatic and induction terms dominate) while they are decisive for, e.g., hydrocarbon or rare gas complexes was questioned for the first time in the years 2001−2002. It is attributed to the growing interest in accurate quantum chemistry based biomolecular simulations for proteins and DNA fragments at that time.115−117 In that time the groups of Yang, Elstner/ Hobza, and Scoles more or less independently published dispersion-corrected DFT approaches (or tight-binding versions) aiming at a more general description of noncovalent interactions in bio-organic systems.118−120 These works did not add anything particularly new from the methodological standpoint because they just applied a damped-dispersion scheme with typical density functionals (nowadays known as DFT-D). However, they marked a turning point in the rediscovery of the problem in DFT and motivated all further more general and widely used DFT-D schemes. The work of Scoles et al.118 made the most general statements regarding the applicability of DFT-D (although this modern term was in fact not used). They tested various functionals and applied it even to transition metal carbonyls. Standard atomic dispersion coefficients, combinatorial rules, and special damping functions were employed, and the use of a three-body dispersion term was already mentioned in this work. Wu and Yang120 considered organic systems and used atomic hybridization state dependent (but not system-specific) dispersion coefficients. Different combinatorial rules and damping functions as well as three typical GGA and one hybrid (B3LYP) functional were tested for RG dimers, for DNA base stacking situations, and even for the intramolecular case of polyalanine conformations. The same basic approach was used two years later by Zimmerli et al.121 to describe water−benzene complexes with B3LYP. The paper of Elstner et al.119 is probably the first where a dispersion correction is employed in a modern semiempirical context. They combined it with the self-consistent-charge, density-functional tight-binding (SCC-DFTB) approach, which represents a minimal-basis set, second-order approximation to a GGA calculation.122 The form of the dispersion correction is very similar to the one used by Wu and Yang. DNA base complexes of stacked and hydrogen-bonded type were investigated in this first publication, and a very good agreement with corresponding MP2 results was reported.

(12)

were proposed, where the ϵm = Em − E0 are atomic excitation energies, the prime implies summation over all states of the atoms except the ground (0) states, and the f ml(A) are the 2l-pole oscillator strengths of the atoms. This GK + dispersion approach was shown to give a reasonable estimate of the whole potential curve for both like and unlike pairs of RG atoms. Simple but seemingly reliable combinatorial rules were also proposed and tested for the higher-order dispersion coefficients. An additional exchange correction according to Rae109 (termed GKR method) was found to improve the agreement with HF data. Because the above dispersion energy was used in an undamped, thus wrong short-range form, parts of the interatomic potential were not considered and the entire work was focused on the long-range interactions. As was common in all these early works, only diatomic RG complexes were investigated. In a series of papers published in the late 1990s,110−113 the group of Gianturco applied a special version of dispersioncorrected DFT to RG(He, Ne, Ar)−CO vdW complexes. The treatment involved a so-called half-and-half density functional of 1 1 LDA global hybrid type (Exc = 2 EXFock + 2 EXC ), a large standard Gaussian AO basis set, and a dispersion term of the form as given in eq 11. The dispersion coefficients were dependent on the bending angle θ in the triatomic complex. Two approaches to merge the DFT and dispersion parts were considered. The first was multiplication of the dispersion potential by a damping function of the Tang−Toennies (TT) type114 n (n) f damp,TT (bR ) = 1 − exp( −bR ) ∑ k=0

(bR )k k!

(13)

where b is an empirical parameter and n defines the order of the dispersion expansion (usually n = 6, 8). This is similar to the treatment in the HFD model of Ahlrichs and Scoles (see below). In modern notation this form of the damping function that asymptotically approaches zero for R → 0 is called “zerodamping” (see section 4 for further discussion). The second approach was somewhat less empirical and involves the shortand medium-range (SMR) correlation energy estimated from the difference of DFT and HF interatomic potentials, i.e., Ecorr,SMR (R ) = E DFT(R ) − E HF(R )

(14)

which was then used to define a total correlation correction, Ecorr,tot(R ) = DλEcorr,SMR (R ) + Edisp(R )

(16)

(15)

3.2. Dispersion-corrected HF

Scoles and co-workers123 started from the pure dispersion interaction in the H2 3Σ+u two-electron system, which is known to a high degree of accuracy, and compared it to the corresponding, entirely repulsive HF potential. It is noted that triplet H2 (not

where Dλ is a scaling parameter obtained by matching the two regions at the points Rλ where the logarithmic derivatives of the dispersion and correlation branches are equal. The final energy was then given by 5112

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

4. MODERN DISPERSION CORRECTIONS

He2) is actually the simplest chemical system showing clear dispersion interactions. From this comparison the authors derived the correction potential needed for the dispersion. They obtained the following relation for the total energy of the dispersion-corrected HF method (often termed HFD) as a function of interatomic distance R, ⎛C C ⎞ C ⎟ E HFD(R ) = E HF − ⎜ 66 + 88 + 10 ⎝R R R10 ⎠ ⎡ ⎛5 ⎞1.9⎤ × exp⎢ −⎜ R m/R − 1⎟ ⎥ ⎠ ⎦ ⎣ ⎝4

4.1. Semiclassical treatments of the dispersion interaction

In semiclassical treatments, the dispersion energy evaluated between atom pairs is simply added to the electronic energy of the mean-field approach (MF; e.g., DFT, HF, or SE-MO) as shown in eq 10. The term “semiclassical” originates from the fact that the dispersion interaction is evaluated as the effective classical interaction between atoms (or molecules), commonly known as London or vdW interaction, although it is intrinsically a purely quantum mechanical interaction. The most widely used semiclassical methods are the exchange-dipole moment (XDM) method by Becke and Johnson,22,83,84 the Tkatchenko−Scheffler86 method, and the D3 approach85,129 (as well as its precursor D2)81 by Grimme and co-workers. Extensions accounting for many-body effects in different flavors have been reported as well.85,87,134 Note that in the current literature some (but not all) of the atom pairwise methods are denoted with the attribute “empirical”, which is inappropriate. In fact, all dispersioncorrection schemes to MF quantum chemical methods described in this Review except vdW-DF contain empirical elements and fitting to external (theoretical or experimental) reference data. The term empirical may suggest inaccuracy, but a simple quality−empiricism relation does not exist here. Hence, the term “empirical” is misleading and should be avoided in this context. We recommend to replace it by a short attribute denoting the basic physical/theoretical foundation of the applied correction (e.g., semiclassical (or, more specifically, atom pairwise), nonlocal density based, effective one-electron based, etc.). All of the semiclassical approaches share more or less the same theoretical foundation, which can be derived from perturbation theory (PT). The approaches then differ in the truncations and approximations introduced. We will briefly demonstrate this by deriving the −C6/R6 term from second-order perturbation theory (PT2). The electron correlation energy up to secondorder in the perturbation after application of the Slater−Condon rules is given by (cf. eq 6)

(17)

Here, the dispersion coefficients C6−C10 for hydrogen and the RG gas atom pairs (to which the method was applied first) were taken from the literature, and the exponential term represents a short-range damping with appropriate atomic or pair-specific radii Rm. The method was later used by the same group with a modified damping term and medium-sized Gaussian AO basis sets, and was even applied to H2−He mixtures.124 For such polyatomic systems, a standard atom pairwise additive approximation Edisp =

∑ Edisp(RAB) AB

(18)

was employed where the sum is over all unique atom pairs in the system and Edisp(RAB) is the corresponding damped pairwise dispersion energy. As noted before, this approach neglects dispersion-energy nonadditivity (type-B effects). An undamped HF + dispersion approach including terms up to C10 was solely applied to the potential energy curve of the neon dimer in 1973 by Conway and Murrell125 (which was one year earlier than the HFD paper of Hepburn and Scoles). This work, although being rather special by focusing on the exchangeinteraction energy and only mentioning the applied dispersion correction in a small paragraph, is to the best of our knowledge the very first HF + dispersion work. The various HF-based methods mentioned could have easily been extended and generalized to arbitrary polyatomic molecules. In particular, this holds for HF compared to DFT because the former was widely used in chemistry and various computer codes were commonly available already in the 1970s and 1980s. However, HFD methods have been used only occasionally to compute small aggregates of aromatic molecules in later decades.126−128 The good accuracy of HF with a general (e.g., D3) dispersion correction for almost arbitrary noncovalent interactions was mentioned for the first time by Grimme et al.129 The very reasonable performance of HF-D3 was later documented by the same group for various complexes34,130 and was recently reconfirmed by Conrad and Gordon.131 These authors furthermore coupled their so-called EFP dispersion model to HF and DFT, leading to the HF-D(EFP) and DFTD(EFP) methods, which seem to perform similar to the corresponding HF-D3 or DFT-D3 methods on a limited number of smaller test complexes.132 To some extent related to HFD is the Hartree−Fock−Clementi−Corongiu method, which scales the two-electron repulsion integrals to include correlation effects and includes an additional long-range dispersion term.133 This reference describes corresponding potential energy curves for naphthalene−RG complexes.

nocc n virt

PT2 Ecorr = −∑ ∑ ij

ab

[(ia|jb) − (ib|ja)]2 ωai + ωbj

(19)

where ij/ab denote occupied/virtual orbitals, respectively. Here and in all following summations, we sum over unique pairs, triples, etc. without explicitly restricting the sum. We use the Mulliken notation for two electron integrals, i.e., 1 (ia|jb) = ∫ dr1 dr2φi(r1)φa(r1) |r − r | φj(r2)φb(r2). ωai is the ex1

2

citation energy corresponding to an electronic excitation from orbital i to a. At this order in the perturbation, two-body correlation effects, leading to pairwise dispersion, are captured. Many-body dispersion effects are obtained if the perturbation series is extended to higher orders in the interelectronic perturbation parameter. It should be mentioned that in the uncoupled approximation the denominator reduces to the respective orbital energy differences, resulting in the MP2 expression for the correlation energy (see eq 6). When the orbitals i and a are localized on fragment A and orbitals j and b are localized on fragment B (eq 19, compare Figure 4), the second (exchange) term in the numerator (ib|ja) will vanish faster (exponentially) in the asymptotic limit (R → ∞). The first term, however, decays slower, and after rearranging 5113

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

interactions, CAB 8 arises from dipole−quadrupole interactions, and CAB arises from dipole−octupole as well as quadrupole− 10 quadrupole interactions. Due to its slower decay, the dipole− dipole term is asymptotically dominating and most important in the long-range regime. For nonspherically symmetric fragments, the expression of the dispersion energy (eq 20) in a multipole expansion also leads to a CAB 7 dispersion coefficient (and higher orders). The CAB 7 arises from a Casimir−Polder integral (see eq 26) with the dipole−quadrupole polarizability on one fragment, either A or B. Its contribution is typically small and an average of the C7 contribution over all equal-weighted orientations is zero, so that it can be omitted in atom−atom models or effectively absorbed into the symmetric higher-order terms (see section 4.4). If the CAB n for fragments A and B are known, the pairwise dispersion energy can easily be calculated. As discussed in section 3, early approaches made use of CAB n coefficients that were specifically obtained for the systems of interest (typically rare gas atoms). In general, however, the CAB n coefficients are unknown for arbitrary fragments, and particularly in the intramolecular case, the definition of such fragments is nontrivial and may become highly artificial. The most straightforward and natural scheme is the fragmentation into atoms.136 Thereby, the coefficient CIJn for interaction between (molecular) fragments I and J can be decomposed into a sum of atom pairwise terms as

the sums, we can express the PT2 correlation energy in the asymptotic limit as PT2 AB = Edisp =− lim Ecorr

R →∞

∑ ∑ i,a∈A j,b∈B

(ia|jb)2 ωai + ωbj

(20)

Thus, in the asymptotic regime, the PT2 correlation energy reduces to the square of Coulomb interactions between the transition densities ia and jb localized on each fragment A and B, respectively. Describing the Coulomb potential in a multipole expansion, the nonvanishing term of lowest order is the dipole− dipole term AB Edisp

≈−

∑ ∑ i ,a∈A j,b∈B

⎛ μ ⃗ ·μ ⃗ − 3(μ ⃗ · eR⃗ )(μ ⃗ · eR⃗ ) ⎞2 1 ia jb ⎜ ia jb ⎟⎟ ωai + ωbj ⎜⎝ R3 ⎠

(21)

Here, μ⃗ia is the electric transition dipole moment (i → a) and eR⃗ is a unit vector along the interfragment axis. For spherically symmetric fragments A and Bwhich is true, e.g., for the free atomsaveraging over all possible orientations of the transition dipole moments on A and B leads to the following formula for the dispersion energy (6) Edisp

3 =− 2

∑ ∑ i,a∈A j,b∈B

|μia |2 |μjb |2 (ωai + ωbj)R6

(22)

As originally demonstrated by Mavroyannis and Stephen,135 the following integral identity can be introduced, ∞ ωaiωbj 1 2 dω = 2 ωai + ωbj π 0 [ωai + ω 2][ωbj2 + ω 2] (23)

CnIJ =

This allows the formulation of methods that are in principle applicable to any system (inter- as well as intramolecular) as long as the atom pairwise dispersion coefficients are known. However, the polarizabilities and consequently the dispersion coefficients of atoms in a molecular environment differ from free-atom ones (type-A effects). Therefore, the dependence of the atomic dispersion coefficients on the molecular environment needs to be taken into account in some way, for which various schemes are in use (see below). Note that the above pairwise additivity assumption is physically motivated by the approximate additivity of polarizability, which has the dimension of volume. The decomposition in eq 27 is valid as long as type-B and type-C nonadditivity effects are not pronounced (see section 2). This is true for most cases in organic or main group chemistry. Nevertheless, it should be mentioned that, e.g., fullerene dimers already exhibit C6 dispersion coefficients, which are larger in magnitude than the plain sum of atom pairwise contributions.137,138 The expression for the dispersion energy (see eq 25 for E(6) disp) is only defined at long range and will lead to singularities when R approaches zero. Consequently, it must be damped at short interatomic distances to avoid artificial overbinding in typical covalent-bonding regimes. The damped atom pairwise dispersion energy is given by

which is used to factorize the energy denominator: 3 πR6

∫0

ωbj|μjb |2 ⎤ ωai|μia |2 ⎤⎡⎢ ⎥ dω ⎢∑ ⎥ ∑ ⎢⎣ i , a ∈ A ωai 2 + ω2 ⎥⎦⎢⎣ j , b ∈ B ωbj 2 + ω2 ⎥⎦

∞⎡

(24)

Using the identity ωai + ω = ωai − (iω) , we can express the dispersion energy via the averaged, isotropic dynamical dipole− dipole polarizability αA.B(iω) of the fragments A,B at imaginary frequency ω: 2

(6) Edisp =−

3 πR6

∫0

2

2

2



αA(iω)αB(iω) dω = −

C6AB R6

(25)

Here, we have used the Casimir−Polder expression of the dispersion coefficient60 C6AB =

3 π

∫0



αA(iω)αB(iω) dω

(27)

A∈I B∈J



(6) =− Edisp

∑ ∑ CnAB

(26)

The dispersion energy, which is intrinsically an electroncorrelation effect, is reduced to local dynamical properties of the individual fragments. While the poles of the dynamical polarizability arise at real excitation frequencies, the polarizability at an imaginary frequency is a mathematically much simpler object. It decays monotonically from the static value α(0) and vanishes at high frequencies α(iω → ∞). This is of high practical relevance as it significantly reduces the required number of grid points in the numerical frequency integration (in eq 26). If higher multipoles are considered in eqs 19−21, similar (10) expressions for E(8) disp, Edisp , etc. are obtained. The integer numbers 6, 8, and 10 denote the power by which the respective AB AB interactions collected within CAB 6 , C8 , and C10 decay with increasing separation R (compare with eq 11). As demonstrated above, the CAB 6 contains all terms arising from dipole−dipole

AB Edisp =

∑ n = 6,8,10,...

(n) Edisp = −∑



AB n = 6,8,10,...

CnAB (n) f Rn damp

(28)

where f(n) damp denotes a typical damping function that normally decays to zero for R → 0. The different damping models will be discussed separately for each dispersion-correction scheme. Before we present these in some detail, we summarize the key points of the semiclassical treatment of Edisp (compare with Figure 7): 5114

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

it still is widely in use147 and constitutes a comprehensible starting point for developing other approaches. In D2, only the two-body, dipole−dipole dispersion energy E(6) disp is evaluated

(1) Within many-body perturbation theory, the order in the perturbation defines the maximum order of fragments considered in the dispersion energy (i.e., how many fragments ABC are at most involved): EAB disp first arises at second order in PT, Edisp appears first at third order, etc. However, it should be noted that higher-order terms involving intrafragment coupling may implicitly be contained within the pairwise, triplewise terms AB (e.g., EABBA corr effects are included within Edisp) if the fragment polarizability captures these couplings. (2) Within each order in the perturbation, the truncation of the multipole expansion of the Coulomb potential defines the considered dispersion terms with different decay behaviors in R, AB i.e., CAB 6 , C8 , etc. (3) The way in which the atom-in-molecules (or solid) dispersion coefficients are obtained (e.g., incorporation of the molecular environment, type-A effects) significantly affects their accuracy. (4) The shape of the damping function is important for a balanced treatment of both the short/medium range correlation (within the chosen MF) and the long-range dispersion. The latter point requires an additional comment: Ideally, the damping function seamlessly interpolates between the two correlation regimes depicted in Figure 2. However, in reality its contribution goes beyond that as it implicitly absorbs errors of the MF approach in that region as well (e.g., in the description of exchange dispersion). In principle, this is true for all dispersion approaches where the dispersion energy is added to the MF energy (see eq 10), i.e., for semiclassical (section 4.1) and similarly for nonlocal density based (section 4.2) corrections. For a dispersion-free functional specifically designed to avoid the double-counting, see ref 139. In the following, only workable, i.e., generally applicable, semiclassical dispersion corrections are presented that have found application in recent years. Therefore, approaches that solely focus on the calculation of dispersion coefficients without coupling them to a MF method will not be considered in the following (some examples can be found in refs 140−145). Although the multipole expansion is the most intuitive way to describe the perturbing Coulomb field, it is ill-defined at short distances, where it formally diverges. Note that the reduction of the dispersion energy to local fragment properties does not necessarily depend on this expansion. It may be expressed in a more general form via the charge density susceptibility χ(r,r′,ω), which describes the response of the charge density at r due to perturbation at r′ with frequency ω, AB Edisp =−

×

1 2π

D2 Edisp = −s6 ∑ AB

χ (ra, r′a, iω)χ (rb, r′b, iω) |ra − rb r′a − r′b|

R6

Fermi fdamp (R )

(30)

fFermi damp (R)

where s6 is a functional-specific scaling parameter. is a Fermi-type damping function that reduces the dispersion energy to zero at small interatomic separations120 Fermi fdamp (R ) =

1 1+e

20[R /(sR R vdW ) − 1]

(31)

where RvdW is the sum of the vdW radii of the two atoms and sR is a scaling parameter. These are estimated from the electron density obtained by restricted-open-shell HF calculations of the free atoms. The CAB 6 dispersion coefficients in eq 30 are obtained as the geometric mean of the respective homoatomic value CAA 6 C6AB =

C6AAC6BB

(32)

The C6AA coefficients are computed using the empirical relationship C6AA = 0.05NIAαA0

(33)

The scaling factor N increases with the row in the periodic table of elements. The static dipole polarizability α0A and first ionization potentials IA are computed with the PBE0148−150 functional. Because D2 does not use dispersion coefficients from the correct Casimir−Polder expression (eq 26), it is regarded as a fully empirical correction. The CAA 6 are element-specific parameters and are available for all elements up to xenon. They do not, however, contain any dependence on the molecular environment, but instead the atomic dispersion coefficients are averaged for different hybridization states of the atom. This “force-field”like character makes it computationally very efficient. Only the geometry of a molecule is required, and the dispersion energy as well as its gradients are easily calculated. However, for example, the CAA 6 for carbons in ethene and ethane are identical, which limits the accuracy of the method. Originally, the method was proposed together with a reparametrized, nonhybrid variant of the B97151 functional (called B97-D).81 It represents the first DFA that has been specifically designed to include a dispersion correction from the very beginning. This mostly avoids electron correlation doublecounting effects, which sometimes appear with standard DFAs (which already describe correlation effects) when they are augmented with dispersion models that additionally describe correlation. As such, the method is still applied. Neglecting the higherorder atom pairwise dispersion coefficients (CAB n , n ≥ 8) works quite well in B97-D because the flexible B97 functional implicitly captures these effects to some extent. The good interplay between D2 and B97 has also been exploited by Chai and HeadGordon in the development of the range-separated hybrid functional ωB97X-D.147 The D2 method represents one of the first semiclassical approaches available for a large part of the periodic table. However, its major deficiency is the missing dependence of the dispersion energy on the molecular environment. Nevertheless, being one of the first dispersion methods that was widely available for (almost) arbitrary systems, this method became extremely popular, with the original paper81 being cited more

∫ dω

∫ dra dr′a drb dr′b

C6AB

(29)

where the integration over ra and rb is restricted to fragments A and B, respectively. From this expression, one can build a bridge to the nonlocal density functionals presented later in section 4.2 as they make local approximations to the charge-density susceptibility χ(r,r′,ω) and thus recover the dispersion interaction by a more complicated Casimir−Polder-type expression. 4.1.1. D2 approach. The D2 method was presented by Grimme in 200681 as a successor of the D1 approach from 2004, which was limited regarding the covered chemical elements.146 Even though the D2 method (often abbreviated as “-D”) is not considered to be state-of-the-art anymore, it is discussed here as 5115

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

than 5000 times.6 D2 has been combined with several MFs like semiempirical methods152−156 and density functionals.157,158 Furthermore, it has been included as an integral component into some newly developed density functionals.147,159 Because of its very simple and manually tunable character, D2 can easily be modified and special variants have been presented.160 In combination with the B3LYP functional,53,54,58,161,162 Civalleri et al. proposed the B3LYP-D* scheme in which the scaling factor s6 is set equal to 1 (originally 1.05)81 and by rescaling RvdW in the damping function (eq 31).163 This modification improved the performance for structures and cohesive energies of molecular crystals. Steinmann, Csonka, and Corminboeuf 164 have presented a variant (known as “dD10” in the literature) that employs the TT damping function and includes higher-order AB multipole terms (i.e., CAB 8 and C10 ) via recursion relations. A recent approach related to D2, i.e., use of fixed dispersion coefficients, is the spherical atom model by Austin et al.,165 which has been proposed together with the so-called APF-D functional. Here, the dispersion coefficients are obtained from the London formula (eq 1) employing precomputed atomic polarizabilities and ionization energies. 4.1.2. D3 approach. In 2010, Grimme and co-workers proposed the D3 approach.85 In contrast to D2, the molecular environment is explicitly taken into account by the empirical concept of fractional coordination numbers (CNs). The coordination number of an atom A in a molecule (or crystal) is given by CNA =

Natoms

∑ B≠A

To derive system-specific dispersion coefficients, the electric dipole polarizability αA(iω) of the free atom as well as for the atom in differently coordinated hydrides as model systems (AmHn and BkHl in eq 35) was computed. All elements (Z ≤ 94) were treated nonempirically by calculations at different imaginary frequencies via time-dependent density functional theory (TD-DFT).141,142,167 A variant of the PBE0 hybrid functional with a modified amount of Fock exchange ax = 3/8 (called PBE38) in an augmented quadruple-ζ (triple-ζ for Z ≥ 87) basis set has been employed. The PBE38 hybrid functional provides accurate dynamic polarizabilities for rare gas atoms and small molecules. It may be less accurate for electronically complicated systems, but all reference calculations were done on relatively simple and small model molecules. A modified Casimir−Polder expression has been used to compute the A B atom pairwise dispersion coefficients CAB 6,ref(CN ,CN ) for atoms A and B in these reference molecules

∫0



(35)

Here, AmHn and BkHl are the reference compounds with atoms A/B having a specific coordination number CN. The contribution of the hydrogen atoms in the reference molecules is subtracted. These reference compounds reflect typical bonding situations for each element and are distinguished by their CNA. In A B D3, the CAB 6,ref(CN ,CN ) values are precomputed for every possible pair combination (both element and CN) and stored. In the actual D3 calculation for a target system, the atom pairwise CAB 6 dispersion coefficient is evaluated as the Gaussian average of the precomputed reference values,

1 1 + e−16(4(RA ,cov + RB,cov)/(3RAB) − 1)

⎤ 1 ⎡ A m Hn n (iω) − α H2(iω)⎥ α ⎢ ⎣ ⎦ m 2 ⎤ 1⎡ l × ⎢α Bk Hl(iω) − α H2(iω)⎥ dω ⎦ k⎣ 2 3 π

AB C6,ref (CNA , CNB) =

(34)

where the covalent radii RA,cov/RB,cov for each element are taken from ref 166. Atoms in different bonding/hybridization situations have different coordination numbers, which is in agreement with chemical intuition. This is illustrated in Figure 8 for carbon, which has a coordination number of roughly 4 in ethane, 3 in ethene, and about 2 in ethyne. This way, the chemical environment (type-A effect) is taken into account only in terms of the molecular geometry.

N

C6AB(CNA ,

B

CN ) =

N

AB ∑i A ∑ j B C6,ref (CNiA , CNBj )Lij N

N

∑i A ∑ j B Lij

(36)

with Lij = e−4[(CN

A

− CNiA)2 + (CNB − CNBj )2 ]

(37)

Therefore, the D3 method requires only the molecular geometry to obtain CNA and CNB as input. For carbon and nitrogen, the dispersion coefficient for homoatomic pairs CAA 6 is plotted as a function of the coordination number in Figure 8. For clarity, we A B AB will drop the lengthy notation CAB 6 (CN ,CN ) in favor of C6 , implying that this is the system-specific dispersion coefficient obtained via eq 36. In D3, dispersion contributions beyond the lowest dipole−dipole order E(6) disp are taken into account as well. The atom pairwise dipole−quadrupole contribution E(8) disp is included with the respective CAB 8 dispersion coefficient, which is computed recursively from

C8AB = 3C6AB Q AQ B

(38)

with QA =

Figure 8. CN-dependent atom pairwise dispersion coefficient for A homoatomic pairs CAA 6 as a function of the coordination number CN . are drawn as vertical sticks in the respective The reference points CAA 6,ref colors (black for carbon, blue for nitrogen). For carbon, the structures of CH, ethyne, ethene, and ethane (corresponding to the coordination numbers 1−4) are depicted as well.

⟨rA4⟩

ZA

⟨rA 4⟩ ⟨rA 2⟩

(39)

⟨rA2⟩

Here, and are multipole-type expectation values derived from atomic densities and ZA is the atomic number of element A. The ratio is thus an element-specific parameter that AB yields good estimates of CAB 8 based on C6 . These established 5116

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

recursion formulas168 give values for the dispersion coefficients of hydrogen with only a few percent errors.169 Explicit higher multipolar terms (e.g., E(10) disp ) are neglected in D3 but absorbed through the s8 scaling factor into the R−8 term. The atom pairwise, two-body dispersion energy in D3 is then given as D3 Edisp = −∑



AB n = 6,8

sn

CnAB (n) f (R ) Rn damp

(40)

The factors sn scale the individual multipolar contributions. The leading term is set to unity (s6 = 1) for most functionals (except for double-hybrid functionals, which already comprise some MP2 correlation). This is mandatory for the correct asymptotic decay of the dispersion energy. The higher order term s8, however, is a parameter that is fitted (along with the parameters in the damping function f(n) damp(R), see below) for each density functional to avoid double counting at the short and medium ranges (see Figure 2). Two variants of D3 exist that employ different damping schemes. The original D3 approach employs a damping function proposed by Chai and Head-Gordon in the context of the ωB97X-D functional147 by which the dispersion energy is damped to zero at short distances, (n) f damp,zero (R ) =

1 1 + 6(R /(sr , nR 0AB))an

Figure 9. Different D3 damping schemes for the E(6) disp contribution of the Kr−Kr dispersion energy. For D3(BJ) (solid black) and D3(0) (dotted gray), the curves are given using the damping parameters for the BLYP functional. The undamped −C6/R6 contribution is given for comparison (dashed black). (9),D3 Edisp = −∑

Rn Rn + (a1R 0 + a 2)n

(RABRACRBC)3

ABC (9) × f damp (RABC ̅ )

(41)

(43)

(9) (RABC Here, the damping scheme in f damp ̅ ) is that of D3(0) (see

where a6 = 14, a8 = 16, and sr,6 and sr,8 are functional-specific parameters. This original D3 scheme will be dubbed D3(0) in the further context. Inspired by the work of Johnson and Becke,83 the rational (Becke−Johnson) damping scheme was combined with the atom pairwise D3 method.129 Here, the dispersion energy is damped to a finite value at short distances (n) f damp,BJ (R ) =

C9ABC(3 cos θa cos θb cos θc + 1)

eq 41) with RABC ̅ being the geometric mean of the atom pairwise distances RAB, RAC, and RBC. It should be noted that the ATM energy in D3 is always computed with the zero-damping function. Because of the faster decay with R, the ATM energy is smaller in magnitude, and hence the potentially nonphysical effect of the zero-damping function is expected to be less important. The dispersion coefficient CABC is approximated as 9 C9ABC ≈ − C6ABC6ACC6BC

(42)

(44)

Initially, it was not clear whether it is reasonable to include the ATM term. Its contribution to the dispersion energy is typically 40 functionals are available for either the D3(0) or the D3(BJ) approach. The D3(0) variant remains in use only for DFA, which already includes medium-range dispersion and in this way avoids double-counting effects better than D3(BJ). In the original publication,85 it was already proposed to additionally compute the three-body dipole−dipole−dipole dispersion energy (also called Axilrod−Teller−Muto term, ATM):170,171

ATM

D3 Edisp

= −∑



AB n = 6,8



∑ s9

sn

CnAB (n) f (R ) Rn damp

C9ABC(3 cos θa cos θb cos θc + 1)

ABC (9) × f damp (RABC ̅ )

(RABRACRBC)3 (45)

Apart from its accuracy (C6 values are, on average, accurate to ∼5%),11 the major advantage of the D3 approach is the computational efficiency. The dispersion energy is evaluated as a sum over all atom pairs (triples). This makes it much more efficient compared to XDM or TS models, particularly in geometry optimizations and harmonic frequency calculations. 5117

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

with the static polarizability α0A of atom A (in its molecular environment) and where CAA 6 is the respective homoatomic dispersion coefficient. For the free atoms, the static polarizabilities α0A,free and dispersion coefficients CAA 6,free are taken from the literature.186 The homoatomic dispersion coefficients for an atom in a molecule CAA 6 , as well as the corresponding static polarizability, are computed via an empirical relationship to the atomic volume,187

Furthermore, no information about the electronic structure is required, and only the molecular geometry is needed as input. This is different from other “true” post-self-consistent-field (SCF) approaches, where a quantum chemical calculation must precede the dispersion energy calculation. Hence, D3 is easily applicable together with semiempirical approaches130,176−179 without sacrificing either the speed of these approaches or the accuracy of the dispersion treatment. Recently, the D3(BJ) approach has been applied together with a quantum mechanically derived force field as well.180 The degree of empiricism is low (three functional dependent parameters), and it is the only atom pairwise method in which the effect of the chemical environment is explicitly taken into account (at least for the model compounds) in the computation of the dynamic polarizabilities. Recently, Schwabe and co-workers181 have modified the D3(BJ) approach (called D3(CSO) for “C-Six Only”), reducing the number of functional specific parameters to just one while retaining roughly the same accuracy. The atom pairwise D3 (independent of the damping function) as well as D3ATM are consistent methods in the sense that all dispersion contributions up to a certain order in the distance dependence (R−8 for D3, R−9 for D3ATM) are taken into account (see section 4.4). The exclusion of the underlying density (or WF) is, however, a disadvantage. Whenever the electron density around an atom in a molecule or solid is significantly different from the one present in the reference compound, larger errors for the dispersion coefficients can be expected. This occurs, for example, in metal complexes where the metal centers often carry significant partial positive charge, i.e., change their oxidation state relative to the reference molecule. In principle, one can increase the space of reference molecules in the C6 calculations for a specific application. In an investigation of ionic surfaces, electrostatically embedded clusters were used for the calculations of dynamic dipole polarizabilities182 for Na and Mg. Likewise, positively charged model compounds were employed as references for organometallic Pd complexes.183 In a study by Saue and co-workers,184 it was shown that, for C6 coefficients, spin−orbit effects are not relevant for most elements. Because scalar relativistic effects are included within the effective core potentials and corresponding basis sets employed in the computation of the D3 C6 coefficients, the atomic dispersion coefficients compare well with four-component relativistic Kohn−Sham results. 4.1.3. Tkatchenko−Scheffler model. In 2009, Tkatchenko and Scheffler presented a density-dependent, atom pairwise dispersion-correction scheme.86 This model (dubbed TS in the following) is restricted to the lowest-order two-body dispersion energy E(6) disp and takes into account the molecular environment via the electron density (see below). The dispersion energy in TS is given as TS Edisp = −∑ AB

C6AB 6

R

Fermi fdamp (R )

αA0 = vAαA0 ,free

(49)

∫ dr wA(r)ρ(r)r 3 vA = ∫ dr ρAfree (r)r 3 wA(r) =

(50)

ρAfree (r) ∑B ρBfree (r)

(51)

where wA(r) is the atomic Hirshfeld partitioning weight for atom A. In the TS method, changes in the polarizability of an atom A due to the molecular environment are estimated by changes in the atomic volume. Once the (homo)atomic data for the atoms in the molecular environment are calculated, the CAB 6 dispersion coefficients between all atoms in the system can be evaluated using eq 47. For the damping function, the vdW radii are modified in a similar way RA = vA1/3RAfree

(52) 189

The free atom vdW radii are taken from the literature, and for the damping function RvdW = RA + RB is used. The dispersion energy is typically overestimated172,174 by the TS model, particularly for highly ionic species.190 This can be alleviated, e.g., by an iterative adjustment of the Hirshfeld partitioning weights.191 The first approach in this direction is the iterative Hirshfeld I scheme191, which was proposed in this context by Bučko et al.190 Here, the total density of the system (ρ(r) in the numerator of eq 50) remains unchanged, and instead, the reference densities ρfree A (r) of the free atoms that enter the Hirshfeld partitioning are iteratively adjusted. In contrast to a free atom, the reference densities ρfree A (r) may then resemble the density of a free ion (with noninteger electron number). This causes changes only in wA(r) (eq 51), however, due to charge preservation not in the denominator of eq 50. These reference densities seem to be more appropriate in highly polar systems than neutral atomic ones.190 Recently, Tkatchenko and coworkers192 presented an alternative approach to this problem. Here the total TS dispersion inclusive energy is computed selfconsistently with respect to changes in the electron density, i.e., the Hirshfeld partition weights wA(r) are kept constrained. The induced changes in the total density ρ(r) in eq 50 improve the unsatisfactory performance for highly polar and ionic systems.192 Overall, the degree of empiricism is comparable to the one in D3. On the one hand, information on the electronic structure is captured by means of the Hirshfeld partitioning. On the other hand, the use of precomputed reference polarizabilities α0A and CAA 6,free keeps the computation of dispersion coefficients simple. These precomputed entities and their mapping to the target system (via Hirshfeld) define the accuracy of the method.

(46)

2C6AAC6BB (αB0/αA0)C6AA + (αA0/αB0)C6BB

(48)

where vA is the ratio between the ef fective volume of atom A in a molecule and the volume of the free atom. This ratio is obtained by a Hirshfeld partitioning,188

The damping function fFermi damp (R) is similar to the one of Wu and Yang120 and in D2 (see eq 31). Within the TS model, the socalled “average-energy” or “Unsöld” approximation is used.2,3,185 That is, the sum over all excitation energies in eq 24 is replaced by an average excitation energy. This way the CAB 6 are approximated from the homoatomic ones via the Slater−Kirkwood formula,4 C6AB =

AA C6AA = vA 2C6,free

(47) 5118

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

Molecular C6 coefficients from TS show a mean absolute relative deviation of 5.5%.86 4.1.4. TS-based many-body dispersion scheme. In 2012, two extensions of the original, nonself-consistent TS method were introduced by Tkatchenko et al.87 The first one is to account for screening of the dispersion interaction due to the chemical environment, termed self-consistent-screening (SCS) approach. Note that the term “screening” here is somewhat different from its normal use in the context of electrostatics because the dispersion interaction does not involve a static Coulomb potential. The screening by the environment is dependent on the frequency of the electromagnetic radiation that transmits the induced electric field (see above) and is reflected in the dynamic polarizability of an atom A. In TS, instead of the exact dynamic dipole polarizability, the Padé approximation193 for the dynamic dipole polarizability is used, αA(iω) ≈

τAB = (1 − fdamp )τAB + fdamp τAB   SR τAB

MBD Edisp =

(53)

AA 4 C6 3 (αA0)2

1 2

3N

∑ p=1

λp −

3 2

N

∑ ωATS + SCS A

(58)

where λp is the pth eigenvalue of the coupling matrix. Note that the second term in eq 58 is a summation over all atoms, because the uncoupled oscillators are isotropic, and, consequently, has a prefactor of three. This expression has some similarities to the random-phase-approximation (RPA) correlation energy,197 which was pointed out by Tkatchenko et al. in 2013.136 The overall procedure is sketched in Figure 10. The MBD model significantly alleviates the TS problem of overestimating the dispersion energy. Excellent performance was reported for lattice energies of organic crystals174 as well as for

where ωA̅ is an ef fective frequency defined as ωA̅ =

(57)

This way, double-counting of dipole interactions is avoided in the dispersion energy. In MBD, a Fermi-type damping function is used for fdamp (see eq 31).196 Diagonalizing the 3N × 3N coupling matrix yields the many-body dispersion energy as the difference between the zero-point energies of the coupled and uncoupled oscillators,

αA0 1 + (ω/ωA̅ )2

LR τAB

(54)

For this purpose, all atoms are represented as a collection of spherical quantum harmonic oscillators (QHOs). Their coupling influences the dynamic polarizability of each atom at a given imaginary frequency iω. The screened polarizability αTS+SCS (iω) A is then obtained by self-consistently solving αATS + SCS(iω) = αATS(iω) + αATS(iω)

SR TS + SCS αB (iω) ∑ τAB B≠A

(55)

with τSR dipole−dipole interaction AB being the short-ranged and αTS A (iω) being the unscreened dynamic polarizability in TS from eq 53. This is done for several imaginary frequencies iω, and the resulting αTS+SCS (iω) are used to compute the screened CAA A 6,SCS for homoatomic pairs by numerical integration of the Casimir− Polder expression (see eq 26). The screened static polarizability α0,SCS is obtained similarly, and both are used to calculate CAB A 6,SCS via eq 47. To compute the many-body dispersion (MBD) energy, a coupled fluctuating dipole model (CFDM)195 is applied. The atoms in the system of interest are represented as a collection of three-dimensional, isotropic quantum harmonic oscillators with eigenfrequencies ωA̅ SCS, i.e., the “effective” frequency, which is obtained via eq 54 by employing both the screened CAA 6,SCS and α0,SCS . Then, the coupling term in the Hamiltonian for the A coupled oscillators is given as 175,194

N

VCFDM =

3

Figure 10. Schematic illustration of how to arrive at the many-body dispersion (MBD) energy based on the Tkatchenko−Scheffler (TS) method. The starting points are the oscillators (eq 54) from the TS model, which are depicted as differently sized and colored circles. First, these are coupled at short range via dipole−dipole interactions. Selfconsistent solution of these coupled oscillators yields a new set of screened oscillators. This is schematically depicted as a change in size and color of initially “overlapping” circles. Note that the isolated oscillator (depicted as the “non-overlapping” circle) is not affected by short-range dipole−dipole coupling and remains unchanged. In the last step, the screened oscillators interact through dipole−dipole coupling at long range, which yields the final many-body dispersion energy. The same scheme is also valid for the related Wannier function based approach (vdW-QHO-WF-SCS-SR).

3

∑ ∑ ∑ χp τpqLRχq A 1), i.e., induced quadrupoles, octupoles, etc. dHigher many-body terms (n > 2, i.e., three-body, four-body, etc.). eParameters for all elements with Z ≤ xx. Methods without element-specific parameters are applicable to all elements (indicated by “all”). fInformation on chemical environment from fractional coordination number. gDipole−quadupole contribution. hThree-body ATM term. iFull many-body summation of coupled dipoles. jDipole−quadupole, quadupole−quadrupole, and dipole−octupole contributions. kThree-body ATM term is available but not used in typical applications lPotentials need specific parametrization (typically H, B, C, N, and O).35,300,301 a

per element). Additionally, all the semiclassical methods use predefined cutoff radii in the short-range damping function together with (one or two) global parameters to adjust the damping to a certain mean-field method. Note that the precalculated C6 coefficients in the D3 scheme are nonempirically computed by time-dependent DFT (TDDFT) but with an empirically adjusted hybrid density functional (PBE38). DCP approaches require specific parametrization for certain elements in combination with a given basis set and density functional. Consequently, their applicability is typically restricted to only a few elements. Nearly all of the presented methods are in some way, directly or indirectly, based on the electron density. Only the D2 and D3 schemes depend solely on the molecular geometry. The direct dependence on the density has the advantage of seamlessly modeling all atoms in a system in their current electronic structure (e.g., oxidation state), which is not possible in purely geometry-dependent approaches. The quality of the density depends on the underlying mean-field approach and may deteriorate if intrinsic errors of the latter become apparent (e.g., charge-delocalization error in semilocal density functionals).382 Because of the use of a minimal basis set and the integral approximations, this is in particular the case for semiempirical methods. In such cases, the D3 scheme can be ideally coupled with semiempirical Hamiltonians without deteriorating the results.176,179,379 Very promising results are obtained in combination with a tight-binding Hamiltonian,178 and zero differential overlap methods have also been successfully combined with D2152−156 and D3.377,379,383 Due to the semiclassical character, D3 can be used without modification in a classical force field.180 The nonlocal density functionals and all C6-based schemes describe the long-range interactions correctly by construction. Only the one-electron based potentials cannot recover the correct −C6/R6 limit. The importance of higher multipoles in the perturbing field and the inclusion of higher many-body interactions in dispersion corrections is under a very active debate. There seems to be evidence that the required order in the multipole expansion depends crucially on the underlying exchange-correlation functional.201 The Minnesota functional

consistently, VV10 calculations are only a few percent slower than uncorrected treatments. The C6-based D2, D3, and TS methods cause practically no additional computation time compared to HF and semilocal DFT methods.81,85,86,380 An efficient analytical derivative of the three-body interaction for the D3 scheme was implemented recently.381 The diagonalization of the dipole-coupling matrix in the MBD scheme is slightly more expensive, especially if nuclear gradients are considered. The computational cost of the exchange-hole integration to generate the XDM dispersion coefficients is substantial, and the XDM method is probably the most costly C6-based dispersion correction (only surpassed by the related dDsC approach).22,213 Particularly for nuclear gradients, approximations to avoid selfconsistent treatments of XDM were presented.217,227 The oneelectron-based potentials are intrinsically inexpensive but can slow down the SCF convergence with respect to the number of SCF cycles.345,346 Because the nonlocal and the Minnesota-type density functionals evaluate the dispersion energy solely from the electron density, no element-specific parameters are introduced and these methods are applicable to all elements in the periodic table. All semiclassical methods, however, rely on elementspecific atomic data. In D3, the element-pair-specific parameters are the precomputed dispersion coefficients,85 TS (and MBD) uses element-specific, precomputed static polarizabilities and homoatomic dispersion coefficients,86 and XDM employs empirical static polarizabilities.210 While the limited availability of element-specific parameters was the natural limitation of early dispersion approaches (see section 3) and D2 was in fact one of the first methods available for a large number of elements, the modern dispersion-correction schemes cover major parts of the periodic table and have basically no practical limitation. All the semiclassical approaches rely on certain tabulated parameters and hence include a certain amount of empiricism. While D3 uses precalculated C6 coefficients for each element in different coordination numbers (on average three reference points per element), TS and MBD use precalculated C6 coefficients and static polarizabilities of the free atom for each element (two values per element) and XDM uses experimental static polarizabilities of the free atom for each element (one parameter 5133

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

family and the double-hybrid functionals384 in particular seem to cover the medium-range correlation regime to a large degree, and because the dispersion correction should be very long range in character, these only require corrections for the leading order dipole term. This is substantially different in intrinsically more repulsive functionals like BLYP or revPBE. It has been reported that some dispersion corrections perform better in combination with these more repulsive functionals.139,260 There is also some consensus that many-body dispersion contributions are important for large and condensed systems. However, their magnitude seems to depend on the specific dispersion model. In the D3 and XDM models,134,172 the three-body contribution typically amounts to 5 Å).396 This is indicated by the performance of the M06-2X functional. Covering dispersion at medium range only, it yields almost the correct association energy for B(C6F5)3/PtBut3 (error of ∼2 kcal/mol), while its deviation from the reference is much larger for B(C6F5)3/PMes3 (∼5 kcal/mol). Including long-range dispersion by means of the D3(0) scheme significantly improves the results (see Table 5). In the discussion of the potential energy curve for the coronene dimer, the importance of the asymptotically correct treatment of dispersion interactions has been emphasized (see above). Here, we have given an additional example that the proper inclusion of dispersion interactions (also at long range) is necessary to accurately describe the thermochemistry even of moderately sized molecules. For the additional consideration of solvation effects and their partial compensation of dispersion corrections, see ref 396 (cf. also section 4.3.3).

residual deviations from the reference for the equilibrium interaction energy are 1−2 kcal/mol (5−10%), which represents the typical small deviation for those kind of semilocal functionals. Note that the estimated error of the reference DLPNOCCSD(T)/CBS* values is about ±0.5 kcal/mol, mainly due to the remaining basis set incompleteness error, is quite significant in this comparison. The M06-2X functional performs considerably better than the standard functionals near the equilibrium structure and underbinds by only 2.5 kcal/mol even without further (long-range) dispersion corrections. This example shows that the empirical adjustment of the large number of parameters in the M06-2X functional (>30) to equilibrium data of small vdW complexes seems to capture the correct physics. In this part of the potential, the electron densities overlap significantly and are sufficiently strongly deformed relative to those of the fragments. However, the situation is different already at slightly larger interfragment distances where M06-2X starts to underbind systematically. Particularly, this holds true for the asymptotic regime, which is already reached at a distance of ∼5 Å where the M06-2X interaction has almost vanished. Although this can be partially cured by adding the D3(0) correction, it is evident from the DLPNO-CCSD(T) data that the combination of D3(BJ) with purely repulsive functionals works better. This conclusion is quite general and is related to the fact that electron-correlation double-counting effects in the medium-range dispersion regime are difficult to avoid with equilibrium-binding functionals of, for example, the M06 type. This effect is also the reason why only the zero-damping variant D3(0) could be coupled reasonably well with M06-2X.129 Recent experience has shown that functionals with intermediate repulsiveness and a coadjustment of the semilocal functional part with the dispersion correction are most effective to get accurately and robustly working MF methods351 (see also refs 139 and 357 for related functional developments). In a very recent study,396 the association of two bimolecular FLPs was investigated. In this example the mostly noncovalently bound complex between the Lewis acid B(C6F5)3 and the Lewis base (PMes3 or PtBut3) is chemically relevant because a specific preorganization and molecular flexibility are required to activate small molecules like H2397 (see Figure 20). A selection of the computed association energies (using the def2-QZVP398,399 AO basis set) from ref 396 is presented in Table 5. According to reference calculations at the DLPNO-CCSD(T)/CBS* level,395 the association energies of both FLPs are very similar. If dispersion-exclusive methods like plain B3LYP or HF are used, positive association energies (unbound complexes) are obtained. The magnitude of these positive energies reflects the repulsive

5.2. Peptide conformation potential

Our next example is the unfolding of the Ace-Ala-Gly-Ala-NMe tetrapeptide from a folded, right-handed α-helix structure to the unfolded β-strand. As starting structures for the folded and unfolded conformation, we used the αR- and βa-geometries, respectively, from ref 400. They were then optimized using the recently proposed composite approach PBEh-3c351 as implemented in the TURBOMOLE suite of programs (version 7.0).289 Five intermediate structures were generated and optimized with constrained Ace-NMe distance. Single-point calculations were performed on these geometries (B3LYP53,54,58,161,162/def2-QZVP398,399) and two dispersioncorrection schemes were applied: the semiclassical D3(BJ)85,129 scheme (see Figure 9) and the density-dependent, nonlocal VV10 correction (NL)89,274 nonself-consistently in a post-SCF manner. The relative energies with respect to the folded conformer are plotted in Figure 21 as a function of the separation between the terminal carbon atoms and compared to high-level reference data at the DLPNO-CCSD(T)/CBS* level (estimated error: ± 0.5 kcal/mol).395 Further comparisons are made with the B3LYP-DCP/6-31+G(2p,2d)35 composite scheme (shortly denoted as DCP in Figure 21) as well as the Minnesota hybrid functional M06-2X.332 According to our calculations, the folded α-helix conformer is more stable than the unfolded β-strand structure (differences from the relative stability reported in refs 400 and 401 may arise from the constrained dihedral angles used therein). Apparently, dispersion stabilizes the compact, folded structure. Without correction, the B3LYP functional leads to an incorrect relative

Table 5. Association Energies (kcal/mol) of B(C6F5)3/PMes3 and B(C6F5)3/PtBut3; If Not Noted Otherwise, the Values Are Taken from Ref 396 and Were Obtained with the def2-QZVP Basis Set

a

method

B(C6F5)3/PMes3

B(C6F5)3/PtBut3

B3LYP B3LYP-D3 B3LYP-NL M06-2Xa M06-2X-D3(0)a HF HF-D3 referencea,b

5.5 −10.5 −11.7 −5.3 −9.0 7.7 −10.2 −10.3

2.8 −11.2 −12.7 −8.4 −11.4 4.2 −11.3 −10.8

This work. bFrom DLPNO-CCSD(T)/CBS* calculation.395 5137

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

is also stronger in the folded than in the unfolded conformation, thus destabilizing the former. The stabilizing effect due to hydrogen bonds, which are present in the folded conformer and mostly captured by B3LYP, is not sufficient for an accurate description of the relative stability. Including dispersion in a semiclassical way (B3LYP-D3) or using the nonlocal VV10 correction (B3LYP-NL) stabilizes the folded conformer, resulting in excellent agreement with the reference data (discrepancy less than 0.15 kcal/mol). Furthermore, it is observed that, for plain B3LYP, the PES is very bumpy, which is not the case for the reference method (DLPNO-CCSD(T)/ CBS*). B3LYP-D3 and B3LYP-NL show excellent mutual agreement, yielding a comparably smooth PES close to the reference. Only one geometry (point 5) is somewhat understabilized for both schemes, and thus, it seems to be caused by a functional specific artifact (B3LYP-DCP/6-31+G(2p,2d) shows a similar kink here). Overall the long-range dispersion correction, the density functional (at short range), and even the interplay at medium range (strongly dependent on the damping function) work remarkably well in both schemes. Comparison with the plain B3LYP data reveals that additive dispersion corrections (following eq 10) and their attachment to the functional provide the necessary physics to describe such an unfolding process in a quantitatively correct manner. The B3LYP-DCP/6-31+G(2p,2d) and M06-2X follow conceptually different approaches (see section 4.3). Both are capable of predicting qualitatively the correct relative stability for the folded and unfolded conformers (points 1 and 7 in Figure 21). However, in both approaches the relative energy of the unfolded conformer is too low, particularly in the DCP scheme. This is likely due to the lack of long-range dispersion part (see Table 4), which seems to contribute more strongly to the

Figure 21. Potential energy curve for unfolding of the tetrapeptide AceAla-Gly-Ala-NMe. For the B3LYP, B3LYP-D3, B3LYP-NL, and M062X calculations, the def2-QZVP basis set was used. DCP represents B3LYP-DCP/6-31+G(2p,2d).35 The reference energies are obtained at the DLPNO-CCSD(T)/CBS*395 level. The geometries of points 1 (folded), 3, 5, and 7 (unfolded) are depicted as well.

stability of the folded compared to the unfolded conformer. The plain B3LYP functional includes Pauli exchange repulsion, which

Figure 22. Lattice energy of the benzene crystal based on constrained volume optimizations (TPSS-D3 level) with single-point evaluations of various dispersion-corrected MF methods. For each method, the cross shows the position of the energy minimum and the arrow indicates the effect of the added dispersion correction. The reference lattice energy refers to a recent CCSD(T) estimate.419 5138

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

three-body contribution in the ATM approximation has only a small effect and decreases the binding by ∼7%. The PES calculated with PBE-TS has a significantly underestimated minimum (too attractive interaction). However, the minimum geometry is reproduced well at the PBE-TS level. All wellestablished semiclassical dispersion corrections, namely, D3,85 MBD,87 and XDM,22,210,211,216 are very close to each other and agree with the reference within its estimated uncertainty (cf. Figure 22a; the errors are smaller than 1 kcal/mol and 1.3%, respectively, for the lattice energy and the unit cell volume). Both the empirical one-electron potentials (M06L331,332 and DFT-DCACP293) and the nonlocal vdW density functional (vdW-DF2251) provide less-accurate potentials. M06L is too repulsive, which can be partially cured with the D3(0) scheme. However, the geometry is too dense and the overall improvement is only minor. BLYP-DCACP and vdW-DF2 yield a reasonable lattice energy close to the reference, but the equilibrium volume is significantly larger as shown in Figure 22b. These results highlight again the importance of a consistent treatment of dispersion interactions in all distance regimes. In condensed systems, the long-range part is especially significant and must not be neglected. An important point in these comparisons has to be emphasized as a number of studies compare calculated lattice energies or equilibrium geometries directly with the measured observables.203,412,420 Especially for organic crystals of small molecules, the impact of zero-point vibrational (ZPV) and thermal contributions to both the sublimation enthalpy and the minimum geometry (on the free energy surface) can be substantial.418,421,422 For instance, the computed hydrogen cyanide mass density of 1.06 g/cm3 and lattice energy of 10.3 kcal/mol (at TPSS-D3/est CBS level) may be misinterpreted as a failure of the dispersion-corrected density functional. However, the errors of 9% and 21% for the mass density and sublimation enthalpy, respectively, diminish to values of only 1% and 5% when ZPV and thermal effects are properly accounted for.423

stabilization of the folded conformer. The comparably similar structure 2 is described extremely well by both approaches. For B3LYP-DCP/6-31+G(2p,2d) all subsequent geometries are systematically calculated to be too stable. M06-2X draws a different picture. While the unfolded and similar structures (7 and 6, respectively) are too stable, the intermediate structures 3 and 4 are computed to be too high in energy. In conclusion, for a correct description of this unfolding process (and this may hold for polypeptides in general), a proper description of dispersion at long range is mandatory. Good agreement with the reference is obtained for both the B3LYP-D3 and the B3LYP-NL approaches, which produce almost identical results. M06-2X performs well, too, with a slight underestimation of the unfolded conformer likely due to the missing long-range dispersion. B3LYP-DCP/6-31+G(2p,2d) yields qualitatively correct relative energies but quantitatively underestimates the long-range dispersion contribution. 5.3. Benzene crystal

To shed light on the long-range behavior of dispersion-corrected MF methods and, in particular, their influence in more dense (condensed) systems, the analysis of a nonpolar, and therefore dispersion energy-dominated, organic crystal is presented. A sufficiently fast and, at the same time, reasonably accurate electronic structure method would be beneficial for the growing field of organic crystal structure prediction.402−404 Crystalline benzene is one of the simplest organic molecular crystals with aromatic π-stacking. It has various energetically close polymorphs405,406 and serves as the prototypical example to test and judge electronic structure methods based on wave function expansions,407−411 dispersion-corrected DFT,173,174,262,381,412−414 and even SE-MO methods.178,179,415 In Figure 22, we show a potential energy surface (PES) of the benzene crystal. It corresponds to constrained volume optimizations around the equilibrium geometry calculated at the TPSS-D3 level. For the various dispersion-corrected MF methods, single-point energies in a converged projectoraugmented wave (PAW)416,417 basis set are evaluated on this PES as calculated with the VASP program.264−266 The experimental sublimation enthalpy and geometry have been back-corrected for zero-point and thermal effects as described in refs 174, 351, and 418 with the corresponding error estimates. The estimated experimental lattice energy is replaced here by a recent CCSD(T) based estimate419 as theoretical reference value. In parts a and b, semiclassical C6-based and in part c oneelectron-based and nonlocal dispersion schemes are shown. The geometric changes of the crystal are depicted in part d. From Figure 22d, one can see that the noncovalent stacking distance is mainly influenced by the volume constraint PES scan, while the covalent bond distances do not change significantly, as expected for a vdW crystal. The plain density functionals PBE148 and BLYP161,162 cannot describe the binding in a qualitatively correct manner. While PBE shows a very shallow minimum, the BLYP PES is purely repulsive. The binding energy is reasonably good with PBE-D2, but the unit cell volume is significantly underestimated. This probably originates from inaccurate C6 coefficients in the old D2 method. The carbon C6 coefficients are actually closer to the carbon C6 within ethyne (compare with Figure 8) and are too high by ∼20% for the benzene crystal. The PBE-D3 potentials are substantially better and agree very well with the reference. For the benzene crystal, both damping variants seem to be appropriate and yield similar results. The

5.4. Standard benchmark sets for noncovalent interactions

Up to this point, we described individual examples specifically selected for showing representative trends and behaviors. To support the above conclusions, we summarize the performance of different methodologies for well-established benchmark sets. We focus on interaction energy benchmarks for purely noncovalently bound systems, because for these standard test sets sufficient data exist for most methods in the literature. Recently, some effort has been devoted to the investigation of dispersion effects on geometric properties including covalent bonds, rotational constants of medium-sized molecules, noncovalent bond distances, and crystal mass densities.24,324,351,424,425 Concerning the energy benchmarks, two different strategies for the compilation of reference data exist. For sufficiently small systems (S22275,426,427 and S66276), high-level wave function based reference calculations are feasible and typically the CCSD(T) level extrapolated to the CBS limit is applied. For larger systems (S12L,198 S30L,95 X23,174,262 and ICE10418), which are the true target of MF methods, one has to rely on experimental data, which requires proper back-correction of, e.g., thermal, vibrational zero-point, and solvation effects.198 For systems of medium size, e.g., for parts of the S12L set, (local) correlated wave function and quantum Monte Carlo methods are nowadays feasible.175 The use of theoretical references is 5139

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

of theoretical noncovalent interaction calculations. Note that the S22 reference values have been revised twice,426,427 and we use the latest published values here. However, because S22 does not include large systems with >50 atoms, it is not appropriate for testing the performance of methods in the asymptotic, large interatomic distance regime (although systems apart from the S22 equilibrium distances were already analyzed in its original publication).275 Furthermore, charged as well as multiple interacting fragments giving rise to various many-body effects are absent in S22. Such systems are contained in the S12L benchmark set of supramolecular complexes,198 which is used here to demonstrate the performance of methods in realistic applications. The S12L reference is based on back-corrected experimental (free) energies.95 The reliability of these estimated gas-phase interaction energies has been confirmed independently by diffusion Monte Carlo175 and DFT-SAPT calculations.428 Moreover, they compare well with DLPNO-CCSD(T)/CBS*395 values (to be published elsewhere). The third set (X23) of (mostly) organic molecular crystals174,262 can be considered as a periodic extension of the S22 where the asymptotic parts of the noncovalent interaction, specifically the dispersion component, may dominate. The benchmark set X23 was compiled by Otero-de-la-Roza and Johnson262 and further refined by Reilly and Tkatchenko.174 Experimental sublimation enthalpies are corrected for zero-point and thermal effects yielding electronic lattice energies that allow convenient benchmarking. The latter study also estimates the impact of anharmonic contributions on the sublimation energy, and we use these values for benchmarking.174 For all three test sets, the same consistent strategy of comparing the data from the MF methods with reference interaction energies was applied: (i) single-point interaction energies for the given reference structures were computed or taken from the literature, (ii) single-particle basis sets were tried to converge to the basis set limit to rule out incompleteness effects as much as possible, and (iii) composite methods or separate dispersion corrections were applied in the standard form as defined in the original publications. Because many dispersioncorrection schemes can be coupled with various MF methods, there exists a plethora of variants that could be tested. To provide a reasonable but still comprehensible picture, we decided to

preferred because no experimental errors occur and geometric, thermal, and various condensed-phase effects are absent. To provide comprehensible benchmark data, we first summarize the test sets and give the corresponding reference values used in the later comparison. We show results for three typical, standard interaction energy benchmark sets. The first is the very well-known and widely used S22 set of Hobza and coworkers,275 comprising small- to medium-sized, mostly organic complexes in their equilibrium structure summarized in Table 6. Table 6. Reference Energies of the S22 Test Set for Noncovalent Interactions As Introduced by Hobza and Coworkers275 with Refined CCSD(T)/CBS(est) Reference Values (ΔE) by Sherrill and Co-workers427 no.a

name

symmetryb

ΔE ± 1%c

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

ammonia dimer water dimer formic acid dimer formamide dimer uracil dimer 2-pyridoxine·2-aminopyridine adenine·thymine methane dimer ethene dimer benzene·methane benzene dimer pyracine dimer uracil dimer indole·benzene adenine·thymine (stack) ethene·ethine benzene·water benzene·ammonia benzene·cyanide benzene dimer indole·benzene (T-shape) phenol dimer

C2h Cs C2h C2h C2h C1 C1 D3d D2d C3 C2h Cs C2 C1 C1 C2v Cs Cs Cs C2v C1 C1

−3.13 −4.99 −18.75 −16.06 −20.64 −16.93 −16.66 −0.53 −1.47 −1.45 −2.65 −4.26 −9.81 −4.52 −11.73 −1.50 −3.28 −2.31 −4.54 −2.72 −5.63 −7.10

a c

Running number according to ref 275. bSymmetry of complex. Binding energy in kcal/mol as given in ref 427 with estimated error.

This set covers hydrogen-bonded as well as typical vdW complexes, and it has become the defacto standard in the field

Table 7. Reference Energies of the S12L Test Set for Supramolecular Host−Guest Complexes As Introduced by Grimme198 with Refined Back-Corrected Experimental Reference Values from Ref 95 no.a

no.b

name

charge

solventc

temperatured

ΔGe

ΔE ± 5%f

1 2 3 4 9 10 17 18 27 28

2a 2b 3a 3b 4a 4b 5a 5b 6a 6b 7a 7b

TCNA@tweezer DCB@tweezer 3c@pincer 3d@pincer C60@catcher C70@catcher GLH@mcycle BQ@mcycle BuNBH4@CB6 PrNH4@CB6 FECP@CB7 ADOH@CB7

0 0 0 0 0 0 0 0 +1 +1 +2 0

CHCl3 CHCl3 CH2Cl2 CH2Cl2 toluene toluene CHCl3 CHCl3 formic acid/H2O formic acid/H2O H2O H2O

298 298 298 298 293 293 298 298 298 298 298 298

−4.2 −1.4 −2.3 −1.3 −5.3 −5.1 −8.3 −3.3 −6.9 −5.7 −21.1 −14.1

−29.0 −20.8 −23.5 −20.3 −28.4 −29.8 −33.4 −23.3 −82.2 −80.1 −132.7g −24.2

21

a Running number according to ref 95. bRunning number according to ref 198. cSolvent of experimental measurement. dTemperature of ΔG measurement in K. eExperimental ΔG in kcal/mol. fBack-corrected ΔE in kcal/mol with estimated error. gReference value identical to procedure in ref 95.

5140

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

Table 8. Reference Energies of the X23 Test Set for Molecular Crystals As Introduced by Otero-de-la-Roza and Johnson262 and Refined by Reilly and Tkatchenko;174 We Use the Back-Corrected References from Ref 174 with Individual Replacements As Indicated Below no.a

no.b

name

space group

Z (unit)c

temperatured

ΔHe

ΔE ± 5%f

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

1 2 3 4 5 6 7 8 9 10 11

cyclohexanedione acetic acid adamantane ammonia anthracene benzene CO2 cyanamide cytosine ethylcarbanate formamide hexamine imidazole naphthalene oxalic acid α oxalic acid α pyrazine pyrazole succinic acid triazine trioxane uracil urea

P21 Pna21 P4̅ 21c P213 P21a Pbca Pa3̅ Pbca P212121 P21̅ P21/n I4̅ 3m P21/c P21/a Pcba P21/c Pmnn P21cn P21/c R3̅ c R3c P21/a P4̅ 21/m

2 4 2 4 2 4 4 8 4 2 4 1 4 2 4 2 2 8 2 6 6 4 2

233 40 188 160 94 218 150 108 298 168 90 100 123 10 298 298 184 108 100 298 103 298 298

19.39 16.24 13.97 7.12 23.46 10.78 5.88 18.05 37.05h 18.81 17.15 18.12 19.45 17.03 22.39 22.38 13.45 17.29 29.42 13.30 13.44 30.87 22.42

21.18 17.40 16.59 8.89 26.94 13.22g 6.50 19.05 38.57i 20.63 18.93 20.60 20.75 19.53 23.01 22.97 14.65 18.57 31.14 14.75 15.87 32.43 24.50

12 13 14 15 16 17 18 19 20 21

a Running number according to ref 174. bRunning number according to ref 262. cNumber of molecules in the primitive unit cell. dTemperature of Xray measurement in K. eExperimental ΔH at T0 = 298 K in kcal/mol. fBack-corrected zero-point exclusive ΔE at T0 = 0 K in kcal/mol with estimated error. gValue replaced by CCSD(T) estimate from ref 419. hValue replaced by current NIST entry registry number 71-30-7.429 iValue adjusted to match updated ΔH.

Table 9. Mean Deviations (MDs), Mean Absolute Deviations (MADs), and Maximum Absolute Deviations (MAXs) in kcal/mol as Well as Mean Absolute Relative Deviations (MARD%) in % of Interaction Energies Calculated with Selected Dispersion-Corrected MF Methods for the Benchmark Sets S22,275 S12L,198 and X23;174,262 Missing Data Are Indicated with Hyphens S22a MD PBE −2.5c semiclassical methods + PBE PBE-D2 0.5c,r PBE-D3o 0.1c,r PBE-TS 0.3r PBE-MBD 0.0r PBE-XDM −0.3r semiclassical methods + hybrid PBE0-MBD 0.1r PBE0-XDM −0.1r o PBE0-D3 0.1c,r PW6B95-D3o −0.2c,r nonlocal density based PBE-NL 0.2r vdW-DF2 −0.4m effective one-electron potentials B3LYP-DCP −0.2n M06L −0.8c M06-2X −0.2c

MAD

S12Lb

MAX

MARD%

MD

MAD

X23p

MAX

MARD%

MD

MAD

MAX

MARD%

2.6

10.1

55.5

−26.5d

26.5

46.1

82.0

−11.6e

11.6

28.4

59.5

0.6 0.5 0.3 0.5 0.5o

1.6 1.7 1.0 1.9 2.7

13.1 9.9 10.8 9.4 9.2

1.2d −2.3d 6.5d,r −0.4h,r −0.9I

1.5 2.3 6.5 1.0 1.5

3.6 9.1 15.1 3.2 3.2

5.2 5.9 20.9 3.4 5.1

1.5e −0.5f 2.3g 1.2g −0.9r,e

1.8 1.2 2.4 1.5 1.1

6.4 3.8 8.5 3.5 3.9

9.5 6.4 12.6 7.9 5.9

0.6 0.5 0.5 0.3

1.8 2.1 1.9 1.2

8.6 7.1 8.7 5.8

1.3h,r 1.0I − 0.0k

1.7 1.5 − 1.6

4.2 3.8 − 3.3

4.7 5.3 − 4.8

0.3g − −0.4j −

0.9 − 1.1 −

2.2 − 3.1 −

5.6 − 6.2 −

0.5 0.5

1.8 2.8

8.3 6.4

2.9d −

3.1 −

7.3 −

10.2 −

− 1.3r,e

− 1.5

− 3.5

− 8.5

0.3 0.8 0.4

0.8 1.7 1.5

5.4 18.1 7.4

− −1.2d −2.3d

− 2.2 2.5

− 5.6 7.5

− 7.7 7.4

5.1r,l −1.9q,r −

5.1 2.3 −

9.6 5.5 −

26.4 11.3 −

a Reference data from Table 6. bReference data from Table 7. cData from ref 283. dData from ref 172. eData from ref 262; oxalic acid α and β replaced by ref 412. fData from ref 173. gData from ref 174. hData from ref 175. IData from ref 201. jData from ref 381. kData from ref 198. lData from ref 352; the number of systems was reduced to 16. mData from ref 88. nData from ref 430. oIncluding the three-body dispersion energy (ATM). pReference data from Table 8. qEnergies computed on PBE-D3 structures. rThis work; individual data for XDM and TS/MBD related methods kindly provided by A. Otero-de-la-Roza, E. Johnson, and A. Tkatchenko.

5141

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

which is 0.78 kcal/mol431 (at the estimated CBS), i.e., worse than for dispersion-corrected MF methods. The situation for the S12L set of large supramolecular complexes is similar to the S22, but the spread of the performance of the methods is generally larger. As noted above, the MAD values are much larger (e.g., 25.8 kcal/mol for plain PBE) while relative deviations are similar to those in the S22 (e.g., 5−10% for well-performing methods that have MAD values of 2−3 kcal/mol for this set). A notable outlier is PBE-TS with an MARD% of 21.9%, but this is cured at the MBD level. Its poor performance has been attributed to the strong contribution of many-body dispersion in supramolecular complexes (or more dense materials in general).175 However, recently it was shown that, within the PBE-XDM model, higher multipole terms (C8 and C10) seem to be more important compared to high-order many-body contributions.201 It was furthermore concluded that PBE clearly represents the best MF method for this set of complexes in combination with XDM, which is different from conclusions for S22/S66 where many MF methods perform similarly. The observation that the asymptotic part of the dispersion treatment is more relevant for the larger S12L complexes than for the S22 is not very obvious from inspection of the data. While the MARD% of PBE increases from 55.5% to 81.7% in S12L, the MD for the London dispersion-devoid M06L functional is similar for S22 and S12L, and for M06-2X the corresponding increase of the MD from −0.2 to −1.7 kcal/mol is only moderate. The results for the X23 solid-state, lattice-energy benchmark set are in between those of S22 and S12L and lead to similar conclusions. The atom pairwise schemes (except PBE-TS) perform very well with almost no systematic deviations, small MAD values of 1−2 kcal/mol, and MARD% of 6−8%. PBE0MBD is within the chemical accuracy of 1 kcal/mol on the X23 set. While PBE0-D3 has a similar (slightly higher) error, the currently best method is the dispersion-corrected screened hybrid HSE06-D3(ATM) with MAD of 0.8 kcal/mol (corresponding to MARD of 4.6%). This highlights again that, in addition to the dispersion-correction scheme, the underlying semilocal density functional is of importance. We want to stress at this point that the interpretation of benchmark data has to be done carefully as averaged trends are discussed. For instance, the CO2 crystal seems to be a special case, where all considered semiclassical models compute an underestimated lattice energy. However, a benchmark set of 10 ice polymorphs (ICE10: from low- to high-density polymorphs)418 confirms some general trends where PBE-TS and PBE-D3 overestimate the lattice energy. This can be partially attributed to the PBE functional, but other combinations (BLYP-D3 and PBE0-D3) show similar trends. The empirically proven consistency of the theoretical description of the interactions between small molecules, e.g., in dimers and in complex interaction networks like the condensed phase, is a very important result. To what extent the lowest achieved MAD in this work for the X23 set of ∼1 kcal/mol for absolute lattice energies transfers to a corresponding accuracy for relative energy differences of crystal polymorphs has to be studied in more detail. A few studies analyzed the polymorphism of glycine or aspirin,432−435 but this analysis has to be done more rigorously for an extended set of data. This is especially important for the in silico crystal structure prediction.402−404 In the 2015 crystal structure prediction blind test, a variety of dispersioncorrected density functionals have been applied, and especially

concentrate on the widely used and well-known PBE functional as the underlying MF scheme for the semiclassical as well as the VV10 dispersion corrections. For comparison, we also show results for four hybrid functional variants (PBE0-MBD, PBE0XDM, PBE0-D3, and PW6B95-D3), two Minnesota functionals (M06L and M06-2X), and B3LYP-DCP. Note that, because of the larger molecular sizes in S12L (up to ∼160 atoms), the average (gas-phase) interaction energy is much larger than that for S22 (44 kcal/mol vs 7.3 kcal/mol; the average interaction energy per molecule in the unit cell for the X23 set is 20.2 kcal/mol). Therefore, in addition to the absolute error measures mean deviation (MD), mean absolute deviation (MAD), and maximum absolute deviation (MAX), we also provide the mean absolute relative deviation (MARD%) from the reference values in Table 9. We use the individual interaction energies as given in the indicated references of Table 9 and recompute the statistical measures according to the reference values from Tables 6, 7, and 8. Missing data were either recomputed in this work or indicated as a footnote. Because extensive experience exists for the very basic S22 set, we will discuss these data first. Note that most of the methods considered here were empirically developed or explicitly fitted to S22 or similar systems, and hence a good performance for the S22 is a more or less necessary but not fully sufficient condition for general applicability. For this benchmark set, MAD values < 0.5 kcal/mol can be considered as satisfactory while values of 0.2−0.3 kcal/mol (i.e., approaching the accuracy of the reference data) indicate excellent performance. As can be seen from the data in the first row of Table 9, dispersion-uncorrected PBE performs rather poorly with an MAD of 2.6 kcal/mol and an MARD% of >50% for S22. The MD value is negative and has about the same absolute value as the MAD, which indicates that almost all interaction energies are underestimated (underbound complexes). Note that PBE belongs to the class of functionals with intermediate repulsiveness, and even worse results (i.e., more strongly underbound complexes) are obtained for uncorrected BLYP, B3LYP, or TPSS (MAD values are 4.8, 3.8, and 3.5 kcal/mol, respectively11). All semiclassical schemes provide good corrections leading to MAD and MARD% values of about 0.5 kcal/mol and 10%, respectively. The nonlocal methods PBE-NL and vdW-DF2 perform similarly, and this also holds for D3-corrected standard functionals (MAD values of 0.29−0.46 kcal/mol; see, e.g., ref 11). The tendency of vdW-DF1 to overestimate dispersion interactions at equilibrium distances diminishes in the revised version, vdW-DF2.233 For comparison, the MAD of vdW-DF1 on the S22 set is 1.44 kcal/mol with revPBE,287 1.03 kcal/mol with revised PW86,287 and 0.23 kcal/ mol with a specifically adapted version of the B88 functional denoted opt-B88,259 and rPW86-VV09 yields an MAD of 1.20 kcal/mol287 with LDA correlation contributing significantly to the binding. In the class of hybrids, PW6B95-D3 yields excellent results, while among the one-electron potential-based methods, B3LYP-DCP stands out. From this analysis and additional results for the related S66 set313 (including many other functionals), one can conclude that various, rather differently constructed dispersion corrections to MF methods provide similarly accurate results for small- and medium-sized noncovalent complexes. The dependence of the underlying functional is roughly the same (or even larger) as that of the various dispersion-correction schemes. The typical error is 5−10% of the interaction energy. To put this into perspective, we mention the MAD on the S22 set obtained for the computationally cheapest post-HF WFT method, MP2, 5142

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

Molecular dispersion effects are then partly quenched, i.e., intramolecular contributions are replaced by intermolecular ones with the solvent. An accurate account of these effects requires sophisticated solvation models with the same high accuracy as DCMF, which is difficult to obtain at present. Despite the versatility and general reliability of modern DCMF, intense research efforts are devoted to advance even further the current frontiers of (i) generality/robustness, (ii) many-body effects, and (iii) accuracy. The situation regarding the last point is relatively clear: often the accuracy of the approximated long-range dispersion energy (typical relative error of 5%) is on an absolute scale higher than that of the underlying mean-field method (i.e., a typical semilocal(hybrid) functional like B3LYP). Hence, further development of accurate short-range exchange-correlation functionals seems warranted before turning back to the dispersion problem. This, however, mostly holds for saturated, electronically localized large gap systems where dispersion is inherently a local phenomenon with small many-body contributions. The latter become more significant in metals or for quasi-metallic situations, and here the errors are larger (but sometimes still acceptable). The question remains open whether the separation of dynamical and static electron correlation is still valid in this case or if DCMF theory is applicable at all. Higher many-body contributions in the nonlocal density functionals via an Axilrod−Teller−Muto-type triple spatial integral is another potential development area in this context. The competitors to DCMF methods are, of course, correlated wave function based approaches like the coupled-cluster “gold standard” CCSD(T), its various local approximations (wave function truncations), and quantum Monte Carlo methods. While these methods are fundamentally more accurate and general, they are still hampered in practice by the slow convergence with the basis set size, related large basis set superposition errors, the huge computational effort involved, and their (technical) inability to efficiently provide the very important nuclear gradients. What is said here about CCSD(T) and related WFTs also partially applies to a number of modern DFT-based methods that are able to directly capture dispersion interactions seamlessly and that have not been described in this review: RPA,29 ab initio DFT,437 SAPT using DFT monomer descriptions,31 double-hybrid functionals,384 or range-separated functionals with a long-range correlation contribution from wave function methods.438 All these methods are often less empirical than dispersion-corrected DFT methods. They may also be generally more applicable and resolve other shortcomings of standard DFT like the self-interaction error or static correlation problems. However, the fact that the correlation functional of these methods depends also on virtual orbitals results in an increased computational cost that severely limits their applicability. Hence, it is clear that, in the foreseeable future, DCMFs will continue to hold their ground, particularly for structure optimization or molecular dynamics treatments. In any case, the most promising way seems to be a combination of DCMF structures with accurate WFT methods in single-point energy mode or WFT-generated reference data to check the DCMF results. Regarding this “third” combined DCMF/WFT way, the future for an accurate electronic structure theory for large or condensed systems is bright.

the newest developments lead to a significant improvement of calculated polymorph landscapes.436 Furthermore, for the pairs of corrections D2 vs D3 as well as TS vs MBD, one notices a significantly improved accuracy for the theoretically better founded method (i.e., D3 and MBD), which is encouraging. Both hybrid functional approaches (PBE0-D3 and PBE0-MBD) lead to further improvement in accuracy (probably due to better EXR/es/ind noncovalent interaction contributions), while vdW-DF2 is on the same level as the PBE/ semiclassical methods. M06L and B3LYP-DCP perform significantly worse by all statistical measures. Particularly, this holds for B3LYP-DCP, which yields quite good results for the small-molecule set S22. Tentatively, this inconsistency can be attributed to the very empirical character of the dispersion correction, which is less transferable to different systems compared to asymptotically correct methods.

6. SUMMARY AND OUTLOOK In the almost 40 years since their first use, corrections for the London (long-range) dispersion energy in mean-field methods like Hartree−Fock or approximate Kohn−Sham DFT have come of age and are now routinely applied in chemistry and physics. Concomitantly to their theoretical and numerical development, we notice an increasing awareness and general acceptance of London dispersion interactions as an important and fruitful general chemical concept. In particular, this involves the new idea of intramolecular dispersion effects and their impact in thermochemistry and for structural problems. The basic reason why it took so long to establish, for example, the widely used DFT-D methods is that the dispersion energy, as a mostly additive quantity, shows up chemically only for relatively large systems (>20−30 atoms). Molecules of this size were simply inaccessible in the early days of quantum chemistry, while nowadays DFT calculations are routinely done on 100−200 atoms and in periodic boundaries even for the condensed phase. In this work, the theoretical foundations of the corrections as well as various approaches have been reviewed. A comprehensive picture of the accuracy of the most widely used methods in applications to typical chemical problems has also been presented. The most prominent correction schemes can be classified into three groups: (i) nonlocal, density-based functionals, (ii) semiclassical C6 based, and (iii) one-electron effective potentials. The properties and pros and cons of these methods were discussed. There seems to be consensus that asymptotic correctness (right −1/R−6 decay behavior with interfragment distance) is a key property for large systems and hence that many methods in group (i) and (ii) yield equally high accuracy for various noncovalent interaction motifs. Furthermore, methods from group (iii) (e.g., highly parametrized density functionals) can be combined with group (i)/(ii) schemes to obtain systematically high accuracy for long- as well as short-range correlation effects. These modern dispersion-corrected mean-field (DCMF) methods are meanwhile widely applied in the chemical and physical community to describe various standard properties like energies or structures. However, there are less well investigated issues like phonon dispersion, elastic and dielectric constants, or excited states that depend on long-range dispersion interactions as well, and this seems to be a promising field that still needs further exploration. A related, but still not completely solved problem that was only briefly discussed in this Review is solvation. Dispersion effects are omnipresent and also occur for any molecule when it is solvated as in most chemical applications. 5143

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

AUTHOR INFORMATION

Roza, A. Tkatchenko, and J. Hoja for helpful discussions and kindly providing reference values concerning the MBD and XDM methods, respectively.

Corresponding Author

*E-mail: [email protected]. Notes

REFERENCES

The authors declare no competing financial interest.

(1) Eisenschitz, R.; London, F. Ü ber das Verhältnis der van der Waalsschen Kräfte zu den homöopolaren Bindungskräften. Eur. Phys. J. A 1930, 60, 491−527. (2) London, F. Zur Theorie und Systematik der Molekularkräfte. Eur. Phys. J. A 1930, 63, 245−279. (3) London, F. The general theory of molecular forces. Trans. Faraday Soc. 1937, 33, 8b−26. (4) Stone, A. J. The Theory of Intermolecular Forces; Oxford University Press: Oxford, U.K., 1997. (5) Kaplan, I. G. Intermolecular Interactions; J. Wiley & Sons: Chichester, U.K., 2006. (6) According to the Web of Science, http://www.webofknowledge. com (accessed December 2, 2015). (7) Gross, E. K. U.; Dobson, J. F.; Petersilka, M. Density functional theory of time-dependent phenomena. Top. Curr. Chem. 1996, 181, 81− 172. (8) Grimme, S.; Antony, J.; Schwabe, T.; Mück-Lichtenfeld, C. Density functional theory with dispersion corrections for supramolecular structures, aggregates, and complexes of (bio)organic molecules. Org. Biomol. Chem. 2007, 5, 741−758. (9) Johnson, E. R.; Mackie, I. D.; DiLabio, G. A. Dispersion interactions in density-functional theory. J. Phys. Org. Chem. 2009, 22, 1127−1135. (10) Burns, L. A.; Vazquez-Mayagoitia, A.; Sumpter, B. G.; Sherrill, C. D. A Comparison of Dispersion Corrections (DFT-D), Exchange-Hole Dipole Moment (XDM) Theory, and Specialized Functionals. J. Chem. Phys. 2011, 134, 084107. (11) Grimme, S. Density functional theory with London dispersion corrections. WIREs Comput. Mol. Sci. 2011, 1, 211−228. (12) Klimes, J.; Michaelides, A. Perspective: Advances and challenges in treating van der Waals dispersion forces in density functional theory. J. Chem. Phys. 2012, 137, 120901. (13) Dobson, J. F.; Gould, T. Calculation of dispersion energies. J. Phys.: Condens. Matter 2012, 24, 073201. (14) Grimme, S. In The Chemical Bond: Chemical Bonding Accross the Periodic Table; Frenking, G., Shaik, S., Eds.; Wiley-VCH: Weinheim, Germany, 2014; pp 477−499. (15) Ikabata, Y.; Nakai, H. Local Response Dispersion Method: A Density-Dependent Dispersion Correction for Density Functional Theory. Int. J. Quantum Chem. 2015, 115, 309−324. (16) Tkatchenko, A. Current Understanding of Van der Waals Effects in Realistic Materials. Adv. Funct. Mater. 2015, 25, 2054−2061. (17) Hellmann, H. Einführung in die Quantenchemie; Franz Deuticke: Leipzig und Wien, 1937. (18) Akimov, A. V.; Prezhdo, O. V. Large-Scale Computations in Chemistry: A Bird’s Eye View of a Vibrant Field. Chem. Rev. 2015, 115, 5797−5890. (19) Grimme, S.; Huenerbein, R.; Ehrlich, S. On the Importance of the Dispersion Energy for the Thermodynamic Stability of Molecules. ChemPhysChem 2011, 12, 1258−1261. (20) Wagner, J. P.; Schreiner, P. R. London Dispersion in Molecular Chemistry-Reconsidering Steric Effects. Angew. Chem., Int. Ed. 2015, 54, 12274−12296. (21) Koide, A. A new expansion for dispersion forces and its application. J. Phys. B: At. Mol. Phys. 1976, 9, 3173. (22) Becke, A. D.; Johnson, E. R. A density-functional model of the dispersion interaction. J. Chem. Phys. 2005, 123, 154101. (23) Riplinger, C.; Sandhoefer, B.; Hansen, A.; Neese, F. Natural triple excitations in local coupled cluster calculations with pair natural orbitals. J. Chem. Phys. 2013, 139, 134101. (24) Grimme, S.; Steinmetz, M. Effects of London Dispersion Correction in Density Functional Theory on the Structures of Organic

Biographies Stefan Grimme studied Chemistry and finished his Ph.D. in 1991 in Physical Chemistry on a topic in laser spectroscopy. He did his habilitation in Theoretical Chemistry in the group of Sigrid Peyerimhoff. In 2000 he got the C4 chair for Theoretical Organic Chemistry at the University of Münster. In 2011 he accepted an offer as the head of the newly founded Mulliken Center for Theoretical Chemistry at the University of Bonn. He has published more than 400 research articles in various areas of quantum chemistry. He is the recipient of the 2013 Schroedinger medal of the World Organization of Theoretically Oriented Chemists (WATOC) and the 2015 Gottfried-WilhelmLeibniz prize of the DFG. His main research interests are the development and application of quantum chemical methods for large molecules, density functional theory, noncovalent interactions, and their impact in chemistry. Andreas Hansen studied chemistry and physics at the TU Chemnitz and the University of Bonn with a special focus on theory. In 2007, he received his diploma degree in chemistry for a thesis about open-shell ab initio single reference methods. Afterwards, he joined the MPI for chemical energy conversion at Mülheim a. d. Ruhr, continued his research on efficient correlated wave function methods, and finished his Ph.D. under the supervision of Frank Neese in 2012. Since then he has been working at the Mulliken Center of Theoretical Chemistry (University of Bonn) as a lecturer and administrative officer in the group of Stefan Grimme. His research interests are the development and benchmarking of wave function and density functional theory methods for large and challenging systems. Jan Gerit Brandenburg received his physics diploma in 2012 from the Institute for Theoretical Physics at Heidelberg University. In 2015 he finished his Ph.D. in theoretical chemistry summa cum laude in the group of Stefan Grimme at University of Bonn, where he stayed half a year as postdoctoral researcher. He is awarded a Feodor-Lynen fellowship of the Humboldt foundation for a joint project with Angelos Michaelides and Sarah L. Price at University College London beginning April 2016. His research concentrates on the development and application of fast electronic structure methods, with particular focus on organic crystals and their polymorph prediction. Christoph Bannwarth received his Master’s degree in chemistry from the RWTH Aachen University in 2012. For his Master’s thesis, he joined the group of Robert W. Woody at the Colorado State University (Fort Collins, U.S.A.). He is currently a third-year Ph.D. student in the group of Stefan Grimme at the University of Bonn. His research deals with the development and application of efficient electronic structure methods in particular with simplified treatments for excited states and related properties.

ACKNOWLEDGMENTS We are grateful to the Deutsche Forschungsgemeinschaft within the SPP 1807 “Control of London dispersion interactions in molecular chemistry” for financial support and to J.-P. Djukic, C. A. Bauer, J. Antony, A. Dunlap-Smith, L. Goerigk, M. Korth, R. Sure, and J. Kubelka for proofreading of the manuscript. Furthermore, we greatly acknowledge S. D. Peyerimhoff for her continuous support. We thank E. R. Johnson, A. Otero-de-la5144

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

Molecules in the Gas Phase. Phys. Chem. Chem. Phys. 2013, 15, 16031− 16042. (25) Ruzsinszky, A.; Perdew, J. P. Twelve outstanding problems in ground-state density functional theory: A bouquet of puzzles. Comput. Theor. Chem. 2011, 963, 2−6. (26) Cohen, A. J.; Mori-Sanchez, P.; Yang, W. Challenges for Density Functional Theory. Chem. Rev. 2012, 112, 289−320. (27) Becke, A. D. Perspective: Fifty years of density-functional theory in chemical physics. J. Chem. Phys. 2014, 140, 18A301. (28) Peverati, R.; Truhlar, D. G. Quest for a universal density functional: the accuracy of density functionals across a broad spectrum of databases in chemistry and physics. Philos. Trans. R. Soc., A 2014, 372, 20120476. (29) Eshuis, H.; Bates, J. E.; Furche, F. Electron correlation methods based on the random phase approximation. Theor. Chem. Acc. 2012, 131, 1084. (30) Szalewicz, K. Symmetry-adapted perturbation theory of intermolecular forces. WIREs Comput. Mol. Sci. 2012, 2, 254−272. (31) Jansen, G. Symmetry-adapted perturbation theory based on density functional theory for noncovalent interactions. WIREs Comput. Mol. Sci. 2014, 4, 127−144. (32) Rajchel, L.; Ż uchowski, P. S.; Szcześniak, M. M.; Chałasiński, G. Density Functional Theory Approach to Noncovalent Interactions via Monomer Polarization and Pauli Blockade. Phys. Rev. Lett. 2010, 104, 163001. (33) Helgaker, T.; Jørgensen, P.; Olsen, J. Molecular Electronic-Structure Theory; J. Wiley: New York, 2000. (34) Kruse, H.; Grimme, S. A geometrical correction for the inter- and intra-molecular basis set superposition error in Hartree-Fock and density functional theory calculations for large systems. J. Chem. Phys. 2012, 136, 154101. (35) DiLabio, G. A.; Johnson, E. R.; Otero-de-la Roza, A. Performance of conventional and dispersion-corrected density-functional theory methods for hydrogen bonding interaction energies. Phys. Chem. Chem. Phys. 2013, 15, 12821−12828. (36) Maurer, S. A.; Lambrecht, D. S.; Kussmann, J.; Ochsenfeld, C. Efficient distance-including integral screening in linear-scaling MøllerPlesset perturbation theory. J. Chem. Phys. 2013, 138, 014101. (37) Schütz, M.; Masur, O.; Usvyat, D. Efficient and accurate treatment of weak pairs in local CCSD(T) calculations. II. Beyond the ring approximation. J. Chem. Phys. 2014, 140, 244107. (38) Eriksen, J. J.; Baudin, P.; Ettenhuber, P.; Kristensen, K.; Kjærgaard, T.; Jørgensen, P. Linear-Scaling Coupled Cluster with Perturbative Triple Excitations: The Divide-Expand-Consolidate CCSD(T) Model. J. Chem. Theory Comput. 2015, 11, 2984−2993. (39) Kállay, M. Linear-scaling implementation of the direct randomphase approximation. J. Chem. Phys. 2015, 142, 204105. (40) Møller, C.; Plesset, M. S. Note on an Approximation Treatment for Many-Electron Systems. Phys. Rev. 1934, 46, 618−622. (41) Jung, Y.; Lochan, R. C.; Dutoi, A. D.; Head-Gordon, M. Scaled opposite-spin second order Møller−Plesset correlation energy: An economical electronic structure method. J. Chem. Phys. 2004, 121, 9793−9802. (42) Cybulski, S. M.; Lytle, M. L. The origin of deficiency of the supermolecule second-order Møller-Plesset approach for evaluating interaction energies. J. Chem. Phys. 2007, 127, 141102. (43) Riley, K. E.; Platts, J. A.; Ř ezác,̌ J.; Hobza, P.; Hill, J. G. Assessment of the Performance of MP2 and MP2 Variants for the Treatment of Noncovalent Interactions. J. Phys. Chem. A 2012, 116, 4159−4169. (44) Sedlak, R.; Janowski, T.; Pitoňaḱ , M.; Ř ezác,̌ J.; Pulay, P.; Hobza, P. Accuracy of Quantum Chemical Methods for Large Noncovalent Complexes. J. Chem. Theory Comput. 2013, 9, 3364−3374. (45) Pitoňaḱ , M.; Heßelmann, A. Accurate Intermolecular Interaction Energies from a Combination of MP2 and TDDFT Response Theory. J. Chem. Theory Comput. 2010, 6, 168−178. (46) Huang, Y.; Goldey, M.; Head-Gordon, M.; Beran, G. J. O. Achieving High-Accuracy Intermolecular Interactions by Combining Coulomb-Attenuated Second-Order Møller-Plesset Perturbation

Theory with Coupled Kohn-Sham Dispersion. J. Chem. Theory Comput. 2014, 10, 2054−2063. (47) Grimme, S. Improved second-order Møller−Plesset perturbation theory by separate scaling of parallel- and antiparallel-spin pair correlation energies. J. Chem. Phys. 2003, 118, 9095−9102. (48) Marchetti, O.; Werner, H.-J. Accurate Calculations of Intermolecular Interaction Energies Using Explicitly Correlated Coupled Cluster Wave Functions and a Dispersion-Weighted MP2Method. J. Phys. Chem. A 2009, 113, 11580−11585. (49) Tkatchenko, A.; DiStasio, R. A.; Head-Gordon, M.; Scheffler, M. Dispersion-corrected Møller−Plesset second-order perturbation theory. J. Chem. Phys. 2009, 131, 094106. (50) Goldey, M. B.; Belzunces, B.; Head-Gordon, M. Attenuated MP2 with a long-range dispersion correction for treating non-bonded interactions. J. Chem. Theory Comput. 2015, 11, 4159−4168. (51) Hättig, C.; Klopper, W.; Köhn, A.; Tew, D. P. Explicitly Correlated Electrons in Molecules. Chem. Rev. 2012, 112, 4−74. (52) Hohenstein, E. G.; Sherrill, C. D. Wavefunction methods for noncovalent interactions. WIREs Comput. Mol. Sci. 2012, 2, 304−326. (53) Becke, A. D. Density-functional thermochemistry. III. The role of exact exchange. J. Chem. Phys. 1993, 98, 5648−5652. (54) Stephens, P. J.; Devlin, F. J.; Chabalowski, C. F.; Frisch, M. J. Ab Initio Calculation of Vibrational Absorption and Circular Dichroism Spectra Using Density Functional Force Fields. J. Phys. Chem. 1994, 98, 11623−11627. (55) Henderson, T. M.; Janesko, B. G.; Scuseria, G. E. Range Separation and Local Hybridization in Density Functional Theory. J. Phys. Chem. A 2008, 112, 12530−12542. (56) Arbuznikov, A. V.; Kaupp, M. Advances in local hybrid exchangecorrelation functionals: from thermochemistry to magnetic-resonance parameters and hyperpolarizabilities. Int. J. Quantum Chem. 2011, 111, 2625−2638. (57) Iikura, H.; Tsuneda, T.; Yanai, T.; Hirao, K. A long-range correction scheme for generalized-gradient-approximation exchange functionals. J. Chem. Phys. 2001, 115, 3540−3544. (58) Becke, A. D. A new mixing of Hartree-Fock and local densityfunctional theories. J. Chem. Phys. 1993, 98, 1372−1377. (59) Csonka, G. I.; Perdew, J. P.; Ruzsinszky, A. Global Hybrid Functionals: A Look at the Engine under the Hood. J. Chem. Theory Comput. 2010, 6, 3688−3703. (60) Casimir, H. B. G.; Polder, D. The Influence of Retardation on the London-van der Waals Forces. Phys. Rev. 1948, 73, 360−372. (61) Davidson, E. R.; Feller, D. Basis set selection for molecular calculations. Chem. Rev. 1986, 86, 681−696. (62) Hirschfelder, J. O. Almost degenerate perturbation theory. Chem. Phys. Lett. 1978, 54, 1−3. (63) Claverie, P. Theory of intermolecular forces. I. On the inadequacy of the usual Rayleigh-Schrö dinger perturbation method for the treatment of intermolecular forces. Int. J. Quantum Chem. 1971, 5, 273−296. (64) Morgan, J. D.; Simon, B. Behavior of molecular potential energy curves for large nuclear separations. Int. J. Quantum Chem. 1980, 17, 1143−1166. (65) Adams, W. H. Perturbation theory of intermolecular interactions: What is the problem, are there solutions? Int. J. Quantum Chem. 1990, 38, 531−547. (66) Kutzelnigg, W. Does the polarization approximation converge for large R to a primitive or a symmetry-adapted wavefunction? Chem. Phys. Lett. 1992, 195, 77−84. (67) Adams, W. H. True or false? Order is not uniquely defined in symmetry adapted perturbation theory. J. Mol. Struct.: THEOCHEM 2002, 591, 59−65. (68) Williams, H. L.; Chabalowski, C. F. Using Kohn-Sham Orbitals in Symmetry-Adapted Perturbation Theory to Investigate Intermolecular Interactions. J. Phys. Chem. A 2001, 105, 646−659. (69) Heßelmann, A.; Jansen, G. First-order intermolecular interaction energies from Kohn-Sham orbitals. Chem. Phys. Lett. 2002, 357, 464− 470. 5145

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(70) Heßelmann, A.; Jansen, G. Intermolecular induction and exchange-induction energies from coupled-perturbed Kohn-Sham density functional theory. Chem. Phys. Lett. 2002, 362, 319−325. (71) Heßelmann, A.; Jansen, G. Intermolecular dispersion energies from time-dependent density functional theory. Chem. Phys. Lett. 2003, 367, 778−784. (72) Misquitta, A. J.; Szalewicz, K. Intermolecular forces from asymptotically corrected density functional description of monomers. Chem. Phys. Lett. 2002, 357, 301−306. (73) Misquitta, A. J.; Jeziorski, B.; Szalewicz, K. Dispersion Energy from Density-Functional Theory Description of Monomers. Phys. Rev. Lett. 2003, 91, 033201. (74) Misquitta, A. J.; Szalewicz, K. Symmetry-adapted perturbationtheory calculations of intermolecular forces employing density-functional description of monomers. J. Chem. Phys. 2005, 122, 214109. (75) Jeziorska, M.; Jeziorski, B.; Č ížek, J. Direct calculation of the Hartree-Fock interaction energy via exchange-perturbation expansion. The He··· He interaction. Int. J. Quantum Chem. 1987, 32, 149−164. (76) Jeziorski, B.; Moszynski, R.; Szalewicz, K. Perturbation Theory Approach to Intermolecular Potential Energy Surfaces of van der Waals Complexes. Chem. Rev. 1994, 94, 1887−1930. (77) Peach, M. J. G.; Williamson, M. J.; Tozer, D. J. Influence of Triplet Instabilities in TDDFT. J. Chem. Theory Comput. 2011, 7, 3578−3585. (78) Ř ezác,̌ J.; Huang, J.; Hobza, P.; Beran, G. J. O. Benchmark Calculations of Three-Body Intermolecular Interactions and the Performance of Low-Cost Electronic Structure Methods. J. Chem. Theory Comput. 2015, 11, 3065−3079. (79) Gordon, M. S.; Slipchenko, L.; Li, H.; Jensen, J. H. The Effective Fragment Potential: A General Method for Predicting Intermolecular Interactions. Annu. Rep. Comput. Chem. 2007, 3, 177−193. (80) Dobson, J. F. Beyond Pairwise Additivity in London Dispersion Interactions. Int. J. Quantum Chem. 2014, 114, 1157−1161. (81) Grimme, S. Semiempirical GGA-type density functional constructed with a long-range dispersion correction. J. Comput. Chem. 2006, 27, 1787−1799. (82) Jurečka, P.; Č erný, j.; Hobza, P.; Salahub, D. R. Density functional theory augmented with an empirical dispersion term. Interaction energies and geometries of 80 noncovalent complexes compared with ab initio quantum mechanics calculations. J. Comput. Chem. 2007, 28, 555− 569. (83) Johnson, E. R.; Becke, A. D. A post-Hartree-Fock model of intermolecular interactions. J. Chem. Phys. 2005, 123, 024101. (84) Johnson, E. R.; Becke, A. D. A post-Hartree-Fock model of intermolecular interactions: Inclusion of higher-order corrections. J. Chem. Phys. 2006, 124, 174104. (85) Grimme, S.; Antony, J.; Ehrlich, S.; Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu. J. Chem. Phys. 2010, 132, 154104. (86) Tkatchenko, A.; Scheffler, M. Accurate Molecular Van Der Waals Interactions from Ground-State Electron Density and Free-Atom Reference Data. Phys. Rev. Lett. 2009, 102, 073005. (87) Tkatchenko, A.; DiStasio, R. A.; Car, R.; Scheffler, M. Accurate and Efficient Method for Many-Body van der Waals Interactions. Phys. Rev. Lett. 2012, 108, 236402. (88) Lee, K.; Murray, É D.; Kong, L.; Lundqvist, B. I.; Langreth, D. C. Higher-accuracy van der Waals density functional. Phys. Rev. B: Condens. Matter Mater. Phys. 2010, 82, 081101. (89) Vydrov, O. A.; Van Voorhis, T. Nonlocal van der Waals density functional: the simpler the better. J. Chem. Phys. 2010, 133, 244103. (90) Tkatchenko, A.; von Lilienfeld, O. A. Popular Kohn-Sham density functionals strongly overestimate many-body interactions in van der Waals systems. Phys. Rev. B: Condens. Matter Mater. Phys. 2008, 78, 045116. (91) von Lilienfeld, O. A.; Tkatchenko, A. Two- and three-body interatomic dispersion energy contributions to binding in molecules and solids. J. Chem. Phys. 2010, 132, 234109.

(92) Dobson, J. F.; White, A.; Rubio, A. Asymptotics of the Dispersion Interaction: Analytic Benchmarks for van der Waals Energy Functionals. Phys. Rev. Lett. 2006, 96, 073201. (93) Due to nonadditivity effects, the C6 dispersion coefficient of the C60 dimer turns out to be too small (i.e., less attractive) by 10−20% if computed by atom pairwise schemes.137,138 However, compared to experimentally derived data, the atom pairwise D3 scheme still yields the correct bonding distance (within 0.1 Å) and energy (within 0.5 kcal/ mol).94 Counterintuitively, the association energy of C60 containing vdW complexes are overestimated (by up to 25%) with atom pairwise D3, which improves if the Axilrod−Teller−Muto term is included (overestimation by ∼10%).95 (94) The potential curve for C60 has been conducted at the TPSSD3(BJ) and BLYP-D3(BJ) in the group of S. Grimme. (95) Sure, R.; Grimme, S. Comprehensive Benchmark of Association (Free) Energies of Realistic Host-Guest Complexes. J. Chem. Theory Comput. 2015, 11, 3785−3801. (96) Gobre, V. V.; Tkatchenko, A. Scaling laws for van der Waals interactions in nanostructured materials. Nat. Commun. 2013, 4, 2341. (97) Tao, J.; Perdew, J. P. Communication: Non-additivity of van der Waals interactions between nanostructures. J. Chem. Phys. 2014, 141, 141101. (98) Kristyán, S.; Pulay, P. Can (semi)local density functional theory account for the London dispersion forces? Chem. Phys. Lett. 1994, 229, 175−180. (99) Pérez-Jordá, J. M.; Becke, A. D. A density-functional study of van der Waals forces: rare gas diatomics. Chem. Phys. Lett. 1995, 233, 134− 137. (100) Hobza, P.; Šponer, J.; Reschel, T. Density functional theory and molecular clusters. J. Comput. Chem. 1995, 16, 1315−1325. (101) Lacks, D. J.; Gordon, R. G. Pair interactions of rare-gas atoms as a test of exchange-energy-density functionals in regions of large density gradients. Phys. Rev. A: At., Mol., Opt. Phys. 1993, 47, 4681−4690. (102) Meijer, E. J.; Sprik, M. A density-functional study of the intermolecular interactions of benzene. J. Chem. Phys. 1996, 105, 8684− 8689. (103) Ziegler, T. Approximate density functional theory as a practical tool in molecular energetics and dynamics. Chem. Rev. 1991, 91, 651− 667. (104) van Mourik, T.; Gdanitz, R. J. A critical note on density functional theory studies on rare-gas dimers. J. Chem. Phys. 2002, 116, 9620−9623. (105) Dalrymple, D. L.; Reinheimer, J. D.; Barnes, D.; Baker, R. London force interactions in the reactions of benzyl chlorides. J. Org. Chem. 1964, 29, 2647−2652. (106) Ghotra, J. S.; Hursthouse, M. B.; Welch, A. J. Three-co-ordinate scandium(III) and europium(III); crystal and molecular structures of their trishexamethyldisilylamides. J. Chem. Soc., Chem. Commun. 1973, 669−670. (107) Cohen, J. S.; Pack, R. T. Modified statistical method for intermolecular potentials. Combining rules for higher van der Waals coefficients. J. Chem. Phys. 1974, 61, 2372−2382. (108) Gordon, R. G.; Kim, Y. S. Theory for the Forces between ClosedShell Atoms and Molecules. J. Chem. Phys. 1972, 56, 3122−3133. (109) Rae, A. I. M. A theory for the interactions between closed shell systems. Chem. Phys. Lett. 1973, 18, 574−577. (110) Gianturco, F. A.; Paesani, F.; Laranjeira, M. F.; Vassilenko, V.; Cunha, M. A.; Shashkov, A. G.; Zolotoukhina, A. F. Computed and measured thermal diffusion factor for CO-He mixtures: a test of recent interaction potentials. Mol. Phys. 1997, 92, 957−972. (111) Gianturco, F. A.; Paesani, F.; Laranjeira, M. F.; Vassilenko, V.; Cunha, M. A.; Shashkov, A. G.; Zolotoukhina, A. F. Computed and measured transport coefficients for CO-He mixtures: testing a density functional approach. Mol. Phys. 1998, 94, 605−622. (112) Gianturco, F. A.; Paesani, F.; Laranjeira, M. F.; Vassilenko, V.; Cunha, M. A. Intermolecular forces from density functional theory. III. A multiproperty analysis for the Ar(1S)-CO(1Σ) interaction. J. Chem. Phys. 1999, 110, 7832−7845. 5146

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(113) Gianturco, F. A.; Paesani, F. The rovibrational structure of the Ar−CO complex from a model interaction potential. J. Chem. Phys. 2001, 115, 249−256. (114) Tang, K. T.; Toennies, J. P. An improved simple model for the van der Waals potential based on universal damping functions for the dispersion coefficients. J. Chem. Phys. 1984, 80, 3726−3741. (115) Seponer, J.; Leszczynski, J.; Hobza, P. Base stacking in cytosine dimer. A comparison of correlated ab initio calculations with three empirical potential models and density functional theory calculations. J. Comput. Chem. 1996, 17, 841−850. (116) Hobza, P.; Šponer, J. Structure, Energetics, and Dynamics of the Nucleic Acid Base Pairs: Nonempirical Ab Initio Calculations. Chem. Rev. 1999, 99, 3247−3276. (117) Müller-Dethlefs, K.; Hobza, P. Noncovalent Interactions: A Challenge for Experiment and Theory. Chem. Rev. 2000, 100, 143−167. (118) Wu, X.; Vargas, M. C.; Nayak, S.; Lotrich, V.; Scoles, G. Towards extending the applicability of density functional theory to weakly bound systems. J. Chem. Phys. 2001, 115, 8748−8757. (119) Elstner, M.; Hobza, P.; Frauenheim, T.; Suhai, S.; Kaxiras, E. Hydrogen bonding and stacking interactions of nucleic acid base pairs: a density-functional-theory based treatment. J. Chem. Phys. 2001, 114, 5149−5155. (120) Wu, Q.; Yang, W. Empirical correction to density functional theory for van der Waals interactions. J. Chem. Phys. 2002, 116, 515− 524. (121) Zimmerli, U.; Parrinello, M.; Koumoutsakos, P. Dispersion corrections to density functionals for water aromatic interactions. J. Chem. Phys. 2004, 120, 2693−2699. (122) Elstner, M.; Porezag, D.; Jungnickel, G.; Elsner, J.; Haugk, M.; Frauenheim, T.; Suhai, S.; Seifert, G. Self-consistent-charge densityfunctional tight-binding method for simulations of complex materials properties. Phys. Rev. B: Condens. Matter Mater. Phys. 1998, 58, 7260− 7268. (123) Hepburn, J.; Scoles, G.; Penco, R. A simple but reliable method for the prediction of intermolecular potentials. Chem. Phys. Lett. 1975, 36, 451−456. (124) Ahlrichs, R.; Penco, R.; Scoles, G. Intermolecular forces in simple systems. Chem. Phys. 1977, 19, 119−130. (125) Conway, A.; Murrell, J. N. The exchange energy between neon atoms. Mol. Phys. 1974, 27, 873−878. (126) Pawliszyn, J.; Szczesniak, M. M.; Scheiner, S. Interactions between aromatic systems: dimers of benzene and s-tetrazine. J. Phys. Chem. 1984, 88, 1726−1730. (127) Č ársky, P.; Selzle, H. L.; Schlag, E. W. Ab initio calculations on the structure of the benzene dimer. Chem. Phys. 1988, 125, 165−170. (128) Gonzalez, C.; Allison, T. C.; Lim, E. C. Hartree-Fock Dispersion Probe of the Equilibrium Structures of Small Microclusters of Benzene and Naphthalene: Comparison with Second-Order Møeller-Plesset Geometries. J. Phys. Chem. A 2001, 105, 10583−10587. (129) Grimme, S.; Ehrlich, S.; Goerigk, L. Effect of the Damping Function in Dispersion Corrected Density Functional Theory. J. Comput. Chem. 2011, 32, 1456−1465. (130) Sure, R.; Grimme, S. Corrected small basis set Hartree-Fock method for large systems. J. Comput. Chem. 2013, 34, 1672−1685. (131) Conrad, J. A.; Gordon, M. S. Modeling Systems with π−π Interactions Using the Hartree-Fock Method with an Empirical Dispersion Correction. J. Phys. Chem. A 2015, 119, 5377−5385. (132) Guidez, E. B.; Gordon, M. S. Dispersion Correction Derived from First Principles for Density Functional Theory and Hartree-Fock Theory. J. Phys. Chem. A 2015, 119, 2161−2168. (133) Clementi, E.; Corongiu, C. Van der Waals Interaction Energies of Helium, Neon, and Argon with Naphthalene. J. Phys. Chem. A 2001, 105, 10379−10383. (134) Otero-de-la-Roza, A.; Johnson, E. R. Many-body dispersion interactions from the exchange-hole dipole moment model. J. Chem. Phys. 2013, 138, 054103. (135) Mavroyannis, C.; Stephen, M. Dispersion forces. Mol. Phys. 1962, 5, 629−638.

(136) Tkatchenko, A.; Ambrosetti, A.; DiStasio, R. A. Interatomic methods for the dispersion energy derived from the adiabatic connection fluctuation-dissipation theorem. J. Chem. Phys. 2013, 138, 074106. (137) Ruzsinszky, A.; Perdew, J. P.; Tao, J.; Csonka, G. I.; Pitarke, J. M. Van der Waals Coefficients for Nanostructures: Fullerenes Defy Conventional Wisdom. Phys. Rev. Lett. 2012, 109, 233203. (138) Kauczor, J.; Norman, P.; Saidi, W. A. Non-additivity of polarizabilities and van der Waals C6 coefficients of fullerenes. J. Chem. Phys. 2013, 138, 114107. (139) Pernal, K.; Podeszwa, R.; Patkowski, K.; Szalewicz, K. Dispersionless Density Functional Theory. Phys. Rev. Lett. 2009, 103, 263201. (140) Stanton, J. F. Calculation of C6 dispersion constants with couyled-cluster theory. Phys. Rev. A: At., Mol., Opt. Phys. 1994, 49, 1698− 1703. (141) van Gisbergen, S. J. A.; Snijders, J. G.; Baerends, E. J. A Density Functional Theory Study of Frequency-Dependent Polarizabilities and van der Waals Dispersion Coefficients for Polyatomic Molecules. J. Chem. Phys. 1995, 103, 9347−9354. (142) Osinga, V. P.; van Gisbergen, S. J. A.; Snijders, J. G.; Baerends, E. J. Density Functional Results for Isotropic and Anisotropic Multipole Polarizabilities and C6, C7 and C8 van der Waals Dispersion Coefficients for Molecules. J. Chem. Phys. 1997, 106, 5091−5101. (143) Cybulski, S. M.; Haley, T. P. New approximations for calculating dispersion coefficients. J. Chem. Phys. 2004, 121, 7711−7716. (144) Tao, J.; Perdew, J. P.; Ruzsinszky, A. Accurate van der Waals coefficients from density functional theory. Proc. Natl. Acad. Sci. U. S. A. 2012, 109, 18−21. (145) Toulouse, J.; Rebolini, E.; Gould, T.; Dobson, J. F.; Seal, P.; Angyán, J. G. Assessment of range-separated time-dependent densityfunctional theory for calculating C6 dispersion coefficients. J. Chem. Phys. 2013, 138, 194106. (146) Grimme, S. Accurate Description of van der Waals Complexes by Density Functional Theory Including Empirical Corrections. J. Comput. Chem. 2004, 25, 1463−1473. (147) Chai, J.-D.; Head-Gordon, M. Long-range corrected hybrid density functionals with damped atom-atom dispersion corrections. Phys. Chem. Chem. Phys. 2008, 10, 6615−6620. (148) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865−3868. (149) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple [Phys. Rev. Lett. 77, 3865 (1996)]. Phys. Rev. Lett. 1997, 78, 1396−1396. (150) Adamo, C.; Barone, V. Toward reliable density functional methods without adjustable parameters: The PBE0 model. J. Chem. Phys. 1999, 110, 6158−6170. (151) Becke, A. D. Density-functional thermochemistry. V. Systematic optimization of exchange-correlation functionals. J. Chem. Phys. 1997, 107, 8554−8560. (152) Collignon, B.; Hoang, P.; Picaud, S.; Liotard, D.; Rayez, M.; Rayez, J. A semi-empirical potential model for calculating interactions between large aromatic molecules and graphite surfaces. J. Mol. Struct.: THEOCHEM 2006, 772, 1−12. (153) McNamara, J. P.; Hillier, I. H. Semi-empirical molecular orbital methods including dispersion corrections for the accurate prediction of the full range of intermolecular interactions in biomolecules. Phys. Chem. Chem. Phys. 2007, 9, 2362−2370. (154) Morgado, C. A.; McNamara, J. P.; Hillier, I. H.; Burton, N. A.; Vincent, M. A. Density functional and semiempirical molecular orbital methods including dispersion corrections for the accurate description of noncovalent interactions involving sulfur-containing molecules. J. Chem. Theory Comput. 2007, 3, 1656−1664. (155) Tuttle, T.; Thiel, W. OM x-D: semiempirical methods with orthogonalization and dispersion corrections. Implementation and biochemical application. Phys. Chem. Chem. Phys. 2008, 10, 2159−2166. (156) Ř ezác,̌ J.; Fanfrlík, J.; Salahub, D.; Hobza, P. Semiempirical Quantum Chemical PM6 Method Augmented by Dispersion and H5147

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

Bonding Correction Terms Reliably Describes Various Types of Noncovalent Complexes. J. Chem. Theory Comput. 2009, 5, 1749−1760. (157) Schwabe, T.; Grimme, S. Double-hybrid density functionals with long-range dispersion corrections: higher accuracy and extended applicability. Phys. Chem. Chem. Phys. 2007, 9, 3397−3406. (158) Yousaf, K. E.; Brothers, E. N. Applications of Screened Hybrid Density Functionals with Empirical Dispersion Corrections to Rare Gas Dimers and Solids. J. Chem. Theory Comput. 2010, 6, 864−872. (159) Swart, M.; Solà, M.; Bickelhaupt, F. M. A new all-round density functional based on spin states and SN2 barriers. J. Chem. Phys. 2009, 131, 094103. (160) Barone, V.; Biczysko, M.; Pavone, M. The role of dispersion correction to DFT for modelling weakly bound molecular complexes in the ground and excited electronic states. Chem. Phys. 2008, 346, 247− 256. (161) Becke, A. D. Density-Functional Exchange-Energy Approximation with Correct Asymptotic Behaviour. Phys. Rev. A: At., Mol., Opt. Phys. 1988, 38, 3098−3100. (162) Lee, C.; Yang, W.; Parr, R. G. Development of the Colle-Salvetti Correlation-Energy Formula into a Functional of the Electron Density. Phys. Rev. B: Condens. Matter Mater. Phys. 1988, 37, 785−789. (163) Civalleri, B.; Zicovich-Wilson, C. M.; Valenzano, L.; Ugliengo, P. B3LYP augmented with an empirical dispersion term (B3LYP-D*) as applied to molecular crystals. CrystEngComm 2008, 10, 405−410. (164) Steinmann, S. N.; Csonka, G.; Corminboeuf, C. Unified Interand Intramolecular Dispersion Correction Formula for Generalized Gradient Approximation Density Functional Theory. J. Chem. Theory Comput. 2009, 5, 2950−2958. (165) Austin, A.; Petersson, G. A.; Frisch, M. J.; Dobek, F. J.; Scalmani, G.; Throssell, K. A Density Functional with Spherical Atom Dispersion Terms. J. Chem. Theory Comput. 2012, 8, 4989−5007. (166) Pyykkö, P.; Atsumi, M. Molecular Single-Bond Covalent Radii for Elements 1−118. Chem. - Eur. J. 2009, 15, 186−197. (167) Casida, M. E. In Recent Advances In Density Functional Methods; Chong, D. P., Ed.; World Scientific: Singapore, 1995; Vol. 1; pp 155− 192. (168) Starkschall, G.; Gordon, R. G. Calculation of Coefficients in the Power Series Expansion of the Long-Range Dispersion Force between Atoms. J. Chem. Phys. 1972, 56, 2801−2806. (169) Thakkar, A. J. Higher dispersion coefficients: Accurate values for hydrogen atoms and simple estimates for other systems. J. Chem. Phys. 1988, 89, 2092−2098. (170) Axilrod, B. M.; Teller, E. Interaction of the van der Waals Type Between Three Atoms. J. Chem. Phys. 1943, 11, 299−300. (171) Muto, Y. Force between nonpolar molecules. Proc. Phys. Soc. Jpn. 1943, 17, 629. (172) Risthaus, T.; Grimme, S. Benchmarking of London DispersionAccounting Density Functional Theory Methods on Very Large Molecular Complexes. J. Chem. Theory Comput. 2013, 9, 1580−1591. (173) Brandenburg, J. G.; Grimme, S. Dispersion Corrected HartreeFock and Density Functional Theory for Organic Crystal Structure Prediction. Top. Curr. Chem. 2013, 345, 1−23. (174) Reilly, A. M.; Tkatchenko, A. Understanding the role of vibrations, exact exchange, and many-body van der Waals interactions in the cohesive properties of molecular crystals. J. Chem. Phys. 2013, 139, 024705. (175) Ambrosetti, A.; Alfè, D.; DiStasio, R. A.; Tkatchenko, A. Hard Numbers for Large Molecules: Toward Exact Energetics for Supramolecular Systems. J. Phys. Chem. Lett. 2014, 5, 849−855. (176) Kromann, J. C.; Christensen, A. S.; Steinmann, C.; Korth, M.; Jensen, J. H. A third-generation dispersion and third-generation hydrogen bonding corrected PM6 method: PM6-D3H+. PeerJ 2014, 2, e449. (177) Grimme, S. Towards First Principles Calculation of Electron Impact Mass Spectra of Molecules. Angew. Chem., Int. Ed. 2013, 52, 6306−6312. (178) Brandenburg, J. G.; Grimme, S. Accurate Modeling of Organic Molecular Crystals by Dispersion-Corrected Density Functional Tight Binding (DFTB). J. Phys. Chem. Lett. 2014, 5, 1785−1789.

(179) Brandenburg, J. G.; Hochheim, M.; Bredow, T.; Grimme, S. Low-Cost Quantum Chemical Methods for Non-Covalent Interactions. J. Phys. Chem. Lett. 2014, 5, 4275−4284. (180) Grimme, S. A General Quantum Mechanically Derived Force Field (QMDFF) for Molecules and Condensed Phase Simulations. J. Chem. Theory Comput. 2014, 10, 4497−4514. (181) Schröder, H.; Creon, A.; Schwabe, T. Reformulation of the D3(Becke-Johnson) Dispersion Correction without Resorting to Higher than C6 Dispersion Coefficients. J. Chem. Theory Comput. 2015, 11, 3163−3170. (182) Ehrlich, S.; Moellmann, J.; Reckien, W.; Bredow, T.; Grimme, S. System-Dependent Dispersion Coefficients for the DFT-D3 Treatment of Adsorption Processes on Ionic Surfaces. ChemPhysChem 2011, 12, 3414−3420. (183) Hansen, A.; Bannwarth, C.; Grimme, S.; Petrović, P.; Werlé, C.; Djukic, J.-P. The Thermochemistry of London Dispersion-Driven Transition Metal Reactions: Getting the ’Right Answer for the Right Reason’. ChemistryOpen 2014, 3, 177−189. (184) Sulzer, D.; Norman, P.; Saue, T. Atomic C 6 dispersion coefficients: a four-component relativistic Kohn−Sham study. Mol. Phys. 2012, 110, 2535−2541. (185) Unsöld, A. Beiträge zur Quantenmechanik der Atome. Ann. Phys. 1927, 387, 355−393. (186) Chu, X.; Dalgarno, A. Linear response time-dependent density functional theory for van der Waals coefficients. J. Chem. Phys. 2004, 121, 4083−4088. (187) Kannemann, F. O.; Becke, A. D. Atomic volumes and polarizabilities in density-functional theory. J. Chem. Phys. 2012, 136, 034109. (188) Hirshfeld, F. Bonded-atom fragments for describing molecular charge densities. Theor. chim. acta 1977, 44, 129−138. (189) Bondi, A. van der Waals volumes and radii. J. Phys. Chem. 1964, 68, 441−451. (190) Bučko, T.; Lebègue, S.; Hafner, J.; Á ngyán, J. G. Improved Density Dependent Correction for the Description of London Dispersion Forces. J. Chem. Theory Comput. 2013, 9, 4293−4299. (191) Bultinck, P.; Van Alsenoy, C.; Ayers, P. W.; Carbó-Dorca, R. Critical analysis and extension of the Hirshfeld atoms in molecules. J. Chem. Phys. 2007, 126, 144111. (192) Ferri, N.; DiStasio, R. A.; Ambrosetti, A.; Car, R.; Tkatchenko, A. Electronic Properties of Molecules and Surfaces with a Self-Consistent Interatomic van der Waals Density Functional. Phys. Rev. Lett. 2015, 114, 176802. (193) Tang, K. T.; Karplus, M. Padé-Approximant Calculation of the Nonretarded van der Waals Coefficients for Two and Three Helium Atoms. Phys. Rev. 1968, 171, 70−74. (194) Silvestrelli, P. L.; Ambrosetti, A. Including screening in van der Waals corrected density functional theory calculations: The case of atoms and small molecules physisorbed on graphene. J. Chem. Phys. 2014, 140, 124107. (195) Donchev, A. G. Many-body effects of dispersion interaction. J. Chem. Phys. 2006, 125, 074713. (196) Ambrosetti, A.; Reilly, A. M.; DiStasio, R. A.; Tkatchenko, A. Long-range correlation energy calculated from coupled atomic response functions. J. Chem. Phys. 2014, 140, 18A508. (197) Langreth, D. C.; Perdew, J. P. Exchange-correlation energy of a metallic surface: Wave-vector analysis. Phys. Rev. B 1977, 15, 2884− 2901. (198) Grimme, S. Supramolecular binding thermodynamics by dispersion corrected density functional theory. Chem. - Eur. J. 2012, 18, 9955−9964. (199) Scuseria, G. E.; Henderson, T. M.; Sorensen, D. C. The ground state correlation energy of the random phase approximation from a ring coupled cluster doubles approach. J. Chem. Phys. 2008, 129, 231101. (200) Scuseria, G. E.; Henderson, T. M.; Bulik, I. W. Particle-particle and quasiparticle random phase approximations: Connections to coupled cluster theory. J. Chem. Phys. 2013, 139, 104113. 5148

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(201) Otero-de-la Roza, A.; Johnson, E. R. Predicting energetics of supramolecular systems using the XDM dispersion model. J. Chem. Theory Comput. 2015, 11, 4033−4040. (202) Blood-Forsythe, M. A.; Markovich, T.; DiStasio, R. A.; Car, R.; Aspuru-Guzik, A. Analytical nuclear gradients for the range-separated many-body dispersion model of noncovalent interactions. Chem. Sci. 2016, 7, 1712−1728. (203) Bučko, T.; Lebègue, S.; Gould, T.; Á ngyán, J. G. Many-body dispersion corrections for periodic systems: an efficient reciprocal space implementation. J. Phys.: Condens. Matter 2016, 28, 045201. (204) Silvestrelli, P. L. Van der Waals Interactions in Density Functional Theory Using Wannier Functions. J. Phys. Chem. A 2009, 113, 5224−5234. (205) Ambrosetti, A.; Silvestrelli, P. L. van der Waals Interactions in density functional theory using Wannier Functions: Improved C6 and C3 coefficients by a different approach. Phys. Rev. B: Condens. Matter Mater. Phys. 2012, 85, 073101. (206) Silvestrelli, P. L. Van der Waals interactions in density functional theory by combining the quantum harmonic oscillator-model with localized Wannier functions. J. Chem. Phys. 2013, 139, 054106. (207) Silvestrelli, P. L. Van der Waals Interactions in DFT Made Easy by Wannier Functions. Phys. Rev. Lett. 2008, 100, 053002. (208) Marzari, N.; Vanderbilt, D. Maximally localized generalized Wannier functions for composite energy bands. Phys. Rev. B: Condens. Matter Mater. Phys. 1997, 56, 12847−12865. (209) Andersson, Y.; Langreth, D. C.; Lundqvist, B. I. van der Waals Interactions in Density-Functional Theory. Phys. Rev. Lett. 1996, 76, 102−105. (210) Becke, A. D.; Johnson, E. R. Exchange-hole dipole moment and the dispersion interaction revisited. J. Chem. Phys. 2007, 127, 154108. (211) Becke, A. D.; Johnson, E. R. Exchange-hole dipole moment and the dispersion interaction. J. Chem. Phys. 2005, 122, 154104. (212) Becke, A. D.; Roussel, M. R. Exchange holes in inhomogeneous systems: A coordinate-space model. Phys. Rev. A: At., Mol., Opt. Phys. 1989, 39, 3761−3767. (213) Kannemann, F. O.; Becke, A. D. Van der Waals Interactions in Density-Functional Theory: Intermolecular Complexes. J. Chem. Theory Comput. 2010, 6, 1081−1088. (214) Cambi, R.; Cappelletti, D.; Liuti, G.; Pirani, F. Generalized correlations in terms of polarizability for van der Waals interaction potential parameter calculations. J. Chem. Phys. 1991, 95, 1852−1861. (215) Lide, D. R., Haynes, W., Eds. CRC Handbook of Chemistry and Physics, 90th ed.; CRC Press: 2010. (216) Becke, A. D.; Johnson, E. R. Exchange-hole dipole moment and the dispersion interaction: High-order dispersion coefficients. J. Chem. Phys. 2006, 124, 014104. (217) Kong, J.; Gan, Z.; Proynov, E.; Freindorf, M.; Furlani, T. Efficient Computation of the Dispersion Interaction with Density-Functional Theory. Phys. Rev. A: At., Mol., Opt. Phys. 2009, 79, 042510. (218) DiLabio, G. A.; Otero-de-la-Roza, A. Noncovalent Interactions in Density-Functional Theory. ArXiv E-Prints 2014; arXiv:1405.1771v2 (accessed on July 23, 2015). (219) Á ngyán, J. G. On the exchange-hole model of London dispersion forces. J. Chem. Phys. 2007, 127, 024108. (220) Heßelmann, A. Derivation of the dispersion energy as an explicit density- and exchange-hole functional. J. Chem. Phys. 2009, 130, 084104. (221) Ayers, P. W. A perspective on the link between the exchange(correlation) hole and dispersion forces. J. Math. Chem. 2009, 46, 86−96. (222) Heßelmann, A. Long-range correlation energies from frequencydependent weighted exchange-hole dipole polarisabilities. J. Chem. Phys. 2012, 136, 014104. (223) Steinmann, S. N.; Corminboeuf, C. A System-Dependent Density-Based Dispersion Correction. J. Chem. Theory Comput. 2010, 6, 1990−2001. (224) Steinmann, S. N.; Corminboeuf, C. A generalized-gradient approximation exchange hole model for dispersion coefficients. J. Chem. Phys. 2011, 134, 044117. (225) Mayer, I.; Salvador, P. Overlap populations, bond orders and valences for ’fuzzy’ atoms. Chem. Phys. Lett. 2004, 383, 368−375.

(226) Steinmann, S. N.; Corminboeuf, C. Comprehensive Benchmarking of a Density-Dependent Dispersion Correction. J. Chem. Theory Comput. 2011, 7, 3567−3577. (227) Corminboeuf, C. Minimizing Density Functional Failures for Non-Covalent Interactions Beyond van der Waals Complexes. Acc. Chem. Res. 2014, 47, 3217−3224. (228) Brémond, E.; Golubev, N.; Steinmann, S. N.; Corminboeuf, C. How important is self-consistency for the dDsC density dependent dispersion correction? J. Chem. Phys. 2014, 140, 18A516. (229) Sato, T.; Nakai, H. Density functional method including weak interactions: Dispersion coefficients based on the local response approximation. J. Chem. Phys. 2009, 131, 224104. (230) Sato, T.; Nakai, H. Local Response Dispersion Method. II. Generalized Multicenter Interactions. J. Chem. Phys. 2010, 133, 194101. (231) Dobson, J. F.; Dinte, B. P. Constraint Satisfaction in Local and Gradient Susceptibility Approximations: Application to a van der Waals Density Functional. Phys. Rev. Lett. 1996, 76, 1780−1783. (232) Becke, A. D. A multicenter numerical integration scheme for polyatomic molecules. J. Chem. Phys. 1988, 88, 2547. (233) Vydrov, O. A.; Van Voorhis, T. Improving the accuracy of the nonlocal van der Waals density functional with minimal empiricism. J. Chem. Phys. 2009, 130, 104105. (234) Giese, T. J.; Audette, V. M.; York, D. M. J. Chem. Phys. 2003, 119, 2618−2622. (235) Kamiya, M.; Tsuneda, T.; Hirao, K. A density functional study of van der Waals interactions. J. Chem. Phys. 2002, 117, 6010−6015. (236) Tsuneda, T.; Suzumura, T.; Hirao, K. A new one-parameter progressive Colle-Salvetti-type correlation functional. J. Chem. Phys. 1999, 110, 10664−10678. (237) Kar, R.; Song, J.-W.; Sato, T.; Hirao, K. Long-range corrected density functionals combined with local response dispersion: A promising method for weak interactions. J. Comput. Chem. 2013, 34, 2353−2359. (238) Alves de Lima, N. Van der Waals density functional from multipole dispersion interactions. J. Chem. Phys. 2010, 132, 014110. (239) Lundqvist, B. I.; Andersson, Y.; Shao, H.; Chan, S.; Langreth, D. Density functional theory including Van Der Waals forces. Int. J. Quantum Chem. 1995, 56, 247−255. (240) Langreth, D.; Lundqvist, B. I.; Chakarova-Käck, S. D.; Cooper, V.; Dion, M.; Hyldgaard, P.; Kelkkanen, A.; Kleis, J.; Kong, L.; Li, S.; et al. A density functional for sparse matter. J. Phys.: Condens. Matter 2009, 21, 084203. (241) Berland, K.; Cooper, V. R.; Lee, K.; Schröder, E.; Thonhauser, T.; Hyldgaard, P.; Lundqvist, B. I. van der Waals forces in density functional theory: a review of the vdW-DF method. Rep. Prog. Phys. 2015, 78, 066501. (242) Zaremba, E.; Kohn, W. Van der Waals interaction between an atom and a solid surface. Phys. Rev. B 1976, 13, 2270. (243) Rapcewicz, K.; Ashcroft, N. Fluctuation attraction in condensed matter: A nonlocal functional approach. Phys. Rev. B: Condens. Matter Mater. Phys. 1991, 44, 4032. (244) Zangwill, A.; Soven, P. Density-functional approach to local-field effects in finite systems: Photoabsorption in the rare gases. Phys. Rev. A: At., Mol., Opt. Phys. 1980, 21, 1561. (245) Andersson, Y.; Rydberg, H. Dispersion Coefficients for van der Waals Complexes, Including C60−C60. Phys. Scr. 1999, 60, 211. (246) Gräfenstein, J.; Cremer, D. An efficient algorithm for the densityfunctional theory treatment of dispersion interactions. J. Chem. Phys. 2009, 130, 124105. (247) Hult, E.; Rydberg, H.; Lundqvist, B. I.; Langreth, D. C. Unified treatment of asymptotic van der Waals forces. Phys. Rev. B: Condens. Matter Mater. Phys. 1999, 59, 4708. (248) Rydberg, H.; Lundqvist, B. I.; Langreth, D. C.; Dion, M. Tractable nonlocal correlation density functionals for flat surfaces and slabs. Phys. Rev. B: Condens. Matter Mater. Phys. 2000, 62, 6997. (249) Rydberg, H.; Dion, M.; Jacobson, N.; Schröder, E.; Hyldgaard, P.; Simak, S.; Langreth, D. C.; Lundqvist, B. I. Van der Waals Density Functional for Layered Structures. Phys. Rev. Lett. 2003, 91, 126402. 5149

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(250) Hyldgaard, P.; Berland, K.; Schröder, E. Interpretation of van der Waals density functionals. Phys. Rev. B: Condens. Matter Mater. Phys. 2014, 90, 075148. (251) Dion, M.; Rydberg, H.; Schröder, E.; Langreth, D. C.; Lundqvist, B. I. Van der Waals Density Functional for General Geometries. Phys. Rev. Lett. 2004, 92, 246401. (252) Dion, M.; Rydberg, H.; Schröder, E.; Langreth, D. C.; Lundqvist, B. I. Erratum: Van der Waals Density Functional for General Geometries [Phys. Rev. Lett. 92, 246401 (2004)]. Phys. Rev. Lett. 2005, 95, 109902. (253) Thonhauser, T.; Cooper, V. R.; Li, S.; Puzder, A.; Hyldgaard, P.; Langreth, D. C. Van der Waals density functional: Self-consistent potential and the nature of the van der Waals bond. Phys. Rev. B: Condens. Matter Mater. Phys. 2007, 76, 125112. (254) Vydrov, O. A.; Wu, Q.; Van Voorhis, T. Self-consistent implementation of a nonlocal van der Waals density functional with a Gaussian basis set. J. Chem. Phys. 2008, 129, 014106. (255) Vydrov, O. A.; Van Voorhis, T. Benchmark Assessment of the Accuracy of Several Van der Waals Density Functionals. J. Chem. Theory Comput. 2012, 8, 1. (256) Román-Pérez, G.; Soler, J. M. Efficient Implementation of a van der Waals Density Functional: Application to Double-Wall Carbon Nanotubes. Phys. Rev. Lett. 2009, 103, 096102. (257) Gulans, A.; Puska, M. J.; Nieminen, R. M. Linear-scaling selfconsistent implementation of the van der Waals density functional. Phys. Rev. B: Condens. Matter Mater. Phys. 2009, 79, 201105. (258) Zhang, Y.; Yang, W. Comment on “Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1998, 80, 890. (259) Klimeš, J.; Bowler, D. R.; Michaelides, A. Chemical accuracy for the van der Waals density functional. J. Phys.: Condens. Matter 2010, 22, 022201. (260) Murray, É D.; Lee, K.; Langreth, D. C. Investigation of Exchange Energy Density Functional Accuracy for Interacting Molecules. J. Chem. Theory Comput. 2009, 5, 2754−2762. (261) Elliott, P.; Burke, K. Non-empirical derivation of the parameter in the B88 exchange functional. Can. J. Chem. 2009, 87, 1485−1491. (262) Otero-de-la-Roza, A.; Johnson, E. R. A Benchmark for NonCovalent Interactions in Solids. J. Chem. Phys. 2012, 137, 054103. (263) Berland, K.; Hyldgaard, P. Exchange functional that tests the robustness of the plasmon description of the van der Waals density functional. Phys. Rev. B: Condens. Matter Mater. Phys. 2014, 89, 035412. (264) Kresse, G.; Hafner, J. Ab initio molecular dynamics for liquid metals. Phys. Rev. B: Condens. Matter Mater. Phys. 1993, 47, 558. (265) Kresse, G.; Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B: Condens. Matter Mater. Phys. 1996, 54, 11169. (266) Kresse, G.; Furthmüller, J. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 1996, 6, 15. (267) Giannozzi, P.; et al. QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials. J. Phys.: Condens. Matter 2009, 21, 395502. (268) Soler, J. M.; Artacho, E.; Gale, J. D.; García, A.; Junquera, J.; Ordejón, P.; Sánchez-Portal, D. The SIESTA method for ab initio orderN materials simulation. J. Phys.: Condens. Matter 2002, 14, 2745. (269) Shao, Y.; et al. Advances in methods and algorithms in a modern quantum chemistry program package. Phys. Chem. Chem. Phys. 2006, 8, 3172−3191. (270) Vydrov, O. A.; Van Voorhis, T. Nonlocal van der Waals Density Functional Made Simple. Phys. Rev. Lett. 2009, 103, 063004. (271) Langreth, D. C.; Lundqvist, B. I. Comment on “Nonlocal Van Der Waals Density Functional Made Simple. Phys. Rev. Lett. 2010, 104, 099303. (272) Vydrov, O. A.; Van Voorhis, T. Vydrov and Van Voorhis Reply. Phys. Rev. Lett. 2010, 104, 099304. (273) Vydrov, O. A.; Van Voorhis, T. Dispersion interactions from a local polarizability model. Phys. Rev. A: At., Mol., Opt. Phys. 2010, 81, 062708.

(274) Hujo, W.; Grimme, S. Performance of the van der Waals Density Functional VV10 and (hybrid)GGA Variants for Thermochemistry and Noncovalent Interactions. J. Chem. Theory Comput. 2011, 7, 3866− 3871. (275) Jurečka, P.; Šponer, J.; Č erný, J.; Hobza, P. Benchmark database of accurate (MP2 and CCSD(T) complete basis set limit) interaction energies of small model complexes, DNA base pairs, and amino acid pairs. Phys. Chem. Chem. Phys. 2006, 8, 1985−1993. (276) Ř ezác,̌ J.; Riley, K. E.; Hobza, P. S66: A Well-balanced Database of Benchmark Interaction Energies Relevant to Biomolecular Structures. J. Chem. Theory Comput. 2011, 7, 2427−2438. (277) Aragó, J.; Ortí, E.; Sancho-García, J. C. Nonlocal van der Waals Approach Merged with Double-Hybrid Density Functionals: Toward the Accurate Treatment of Noncovalent Interactions. J. Chem. Theory Comput. 2013, 9, 3437−3443. (278) Tran, F.; Hutter, J. Nonlocal van der Waals functionals: The case of rare-gas dimers and solids. J. Chem. Phys. 2013, 138, 204103. (279) Björkman, T.; Gulans, A.; Krasheninnikov, A. V.; Nieminen, R. M. van der Waals Bonding in Layered Compounds from Advanced Density-Functional First-Principles Calculations. Phys. Rev. Lett. 2012, 108, 235502. (280) Björkman, T.; Gulans, A.; Krasheninnikov, A.; Nieminen, R. Are we van der Waals ready? J. Phys.: Condens. Matter 2012, 24, 424218. (281) Björkman, T. van der Waals density functional for solids. Phys. Rev. B: Condens. Matter Mater. Phys. 2012, 86, 165109. (282) Hujo, W.; Grimme, S. Comparison of the performance of dispersion-corrected density functional theory for weak hydrogen bonds. Phys. Chem. Chem. Phys. 2011, 13, 13942−13950. (283) Goerigk, L.; Grimme, S. Efficient and Accurate Double-HybridMeta-GGA Density Functionals-Evaluation with the Extended GMTKN30 Database for General Main Group Thermochemistry, Kinetics, and Noncovalent Interactions. J. Chem. Theory Comput. 2011, 7, 291−309. (284) Zhao, Y.; Truhlar, D. G. Design of Density Functionals That Are Broadly Accurate for Thermochemistry, Thermochemical Kinetics, and Nonbonded Interactions. J. Phys. Chem. A 2005, 109, 5656−5667. (285) Sperger, T.; Sanhueza, I. A.; Kalvet, I.; Schoenebeck, F. Computational Studies of Synthetically Relevant Homogeneous Organometallic Catalysis Involving Ni, Pd, Ir, and Rh: An Overview of Commonly Employed DFT Methods and Mechanistic Insights. Chem. Rev. 2015, 115, 9532−9586. (286) Petrović, P.; Djukic, J.-P.; Hansen, A.; Bannwarth, C.; Grimme, S. Non-covalent stabilization in transition metal coordination and organometallic complexes. In Non-covalent Interactions in Synthesis and Design of New Compounds; Maharromov, A. M., Mahmudov, K. T., Kopylovich, M. N., Eds.; Wiley-VCH: Weinheim, Germany, 2016; Chapter 7, ISBN 9781119109891, in press. (287) Vydrov, O. A.; Van Voorhis, T. Implementation and assessment of a simple nonlocal van der Waals density functional. J. Chem. Phys. 2010, 132, 164113. (288) Neese, F. The ORCA program system. WIREs Comput. Mol. Sci. 2012, 2, 73−78. (289) Furche, F.; Ahlrichs, R.; Hättig, C.; Klopper, W.; Sierka, M.; Weigend, F. Turbomole. WIREs Comput. Mol. Sci. 2014, 4, 91−100. (290) Feynman, R. P. Forces in Molecules. Phys. Rev. 1939, 56, 340− 343. (291) Hunt, K. L. C. Dispersion dipoles and dispersion forces: Proof of Feynman’s ”conjecture” and generalization to interacting molecules of arbitrary symmetry. J. Chem. Phys. 1990, 92, 1180−1187. (292) Odbadrakh, T. T.; Jordan, K. D. Dispersion dipoles for coupled Drude oscillators. J. Chem. Phys. 2016, 144, 034111. (293) von Lilienfeld, O. A.; Tavernelli, I.; Röthlisberger, U.; Sebastiani, D. Optimization of Effective Atom Centered Potentials for London Dispersion Forces in Density Functional Theory. Phys. Rev. Lett. 2004, 93, 153004. (294) von Lilienfeld, O. A.; Tavernelli, I.; Rothlisberger, U.; Sebastiani, D. Performance of optimized atom-centered potentials for weakly bonded systems using density functional theory. Phys. Rev. B: Condens. Matter Mater. Phys. 2005, 71, 195119. 5150

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(295) Tapavicza, E.; Lin, I.-C.; von Lilienfeld, O. A.; Tavernelli, I.; Coutinho-Neto, M. D.; Rothlisberger, U. Weakly Bonded Complexes of Aliphatic and Aromatic Carbon Compounds Described with Dispersion Corrected Density Functional Theory. J. Chem. Theory Comput. 2007, 3, 1673−1679. (296) Lin, I.-C.; von Lilienfeld, O. A.; Coutinho-Neto, M. D.; Tavernelli, I.; Rothlisberger, U. Predicting Noncovalent Interactions between Aromatic Biomolecules with London-Dispersion-Corrected DFT. J. Phys. Chem. B 2007, 111, 14346−14354. (297) Arey, J. S.; Aeberhard, P. C.; Lin, I.-C.; Rothlisberger, U. Hydrogen Bonding Described Using Dispersion-Corrected Density Functional Theory. J. Phys. Chem. B 2009, 113, 4726−4732. (298) Tkatchenko, A.; von Lilienfeld, O. A. Adsorption of Ar on graphite using London dispersion forces corrected Kohn-Sham density functional theory. Phys. Rev. B: Condens. Matter Mater. Phys. 2006, 73, 153406. (299) Sun, Y. Y.; Kim, Y.-H.; Lee, K.; Zhang, S. B. Accurate and efficient calculation of van der Waals interactions within density functional theory by local atomic potential approach. J. Chem. Phys. 2008, 129, 154102. (300) Mackie, I. D.; DiLabio, G. A. Accurate dispersion interactions from standard density-functional theory methods with small basis sets. Phys. Chem. Chem. Phys. 2010, 12, 6092−8. (301) Torres, E.; DiLabio, G. A. A (Nearly) Universally Applicable Method for Modeling Noncovalent Interactions Using B3LYP. J. Phys. Chem. Lett. 2012, 3, 1738−1744. (302) DiLabio, G. A.; Koleini, M.; Torres, E. Extension of the B3LYPdispersion-correcting potential approach to the accurate treatment of both inter- and intra-molecular interactions. Theor. Chem. Acc. 2013, 132, 1389. (303) van Santen, J. A.; DiLabio, G. A. Dispersion Corrections Improve the Accuracy of Both Noncovalent and Covalent Interactions Energies Predicted by a Density-Functional Theory Approximation. J. Phys. Chem. A 2015, 119, 6703−6713. (304) Andrae, D.; Häußermann, U.; Dolg, M.; Stoll, H.; Preuss, H. Energy-adjusted ab initio pseudopotentials for the second and third row transition elements. Theor. Chim. Acta 1990, 77, 123−141. (305) Peterson, K. A.; Figgen, D.; Goll, E.; Stoll, H.; Dolg, M. Systematically convergent basis sets with relativistic pseudopotentials. II. Small-core pseudopotentials and correlation consistent basis sets for the post-d group 16−18 elements. J. Chem. Phys. 2003, 119, 11113− 11123. (306) Perdew, J. P.; Zunger, A. Self-interaction correction to densityfunctional approximations for many-electron systems. Phys. Rev. B: Condens. Matter Mater. Phys. 1981, 23, 5048−5079. (307) Baumeier, B.; Krüger, P.; Pollmann, J. Self-interaction-corrected pseudopotentials for silicon carbide. Phys. Rev. B: Condens. Matter Mater. Phys. 2006, 73, 195205. (308) Segev, D.; Janotti, A.; Van de Walle, C. G. Self-consistent bandgap corrections in density functional theory using modified pseudopotentials. Phys. Rev. B: Condens. Matter Mater. Phys. 2007, 75, 035201. (309) Lin, I.-C.; Coutinho-Neto, M. D.; Felsenheimer, C.; von Lilienfeld, O. A.; Tavernelli, I.; Rothlisberger, U. Library of dispersioncorrected atom-centered potentials for generalized gradient approximation functionals: Elements H, C, N, O, He, Ne, Ar, and Kr. Phys. Rev. B: Condens. Matter Mater. Phys. 2007, 75, 205131. (310) Aeberhard, P. C.; Arey, J. S.; Lin, I.-C.; Rothlisberger, U. Accurate DFT Descriptions for Weak Interactions of Molecules Containing Sulfur. J. Chem. Theory Comput. 2009, 5, 23−28. (311) Almbladh, C.-O.; von Barth, U. Exact results for the charge and spin densities, exchange-correlation potentials, and density-functional eigenvalues. Phys. Rev. B: Condens. Matter Mater. Phys. 1985, 31, 3231− 3244. (312) Katriel, J.; Davidson, E. R. Asymptotic behavior of atomic and molecular wave functions. Proc. Natl. Acad. Sci. U. S. A. 1980, 77, 4403− 4406.

(313) Goerigk, L.; Kruse, H.; Grimme, S. Benchmarking Density Functional Methods against the S66 and S66 × 8 Datasets for NonCovalent Interactions. ChemPhysChem 2011, 12, 3421−3433. (314) Xu, X.; Goddard, W. A. The X3LYP extended density functional for accurate descriptions of nonbond interactions, spin states, and thermochemical properties. Proc. Natl. Acad. Sci. U. S. A. 2004, 101, 2673−2677. (315) Xu, X.; Goddard, W. A. Bonding Properties of the Water Dimer: A Comparative Study of Density Functional Theories. J. Phys. Chem. A 2004, 108, 2305−2313. (316) Č erný, J.; Hobza, P. The X3LYP extended density functional accurately describes H- bonding but fails completely for stacking. Phys. Chem. Chem. Phys. 2005, 7, 1624−1626. (317) Walsh, T. R. Exact exchange and Wilson−Levy correlation: a pragmatic device for studying complex weakly-bonded systems. Phys. Chem. Chem. Phys. 2005, 7, 443−451. (318) Tao, J.; Perdew, J. P.; Staroverov, V. N.; Scuseria, G. E. Climbing the Density Functional Ladder: Nonempirical Meta Generalized Gradient Approximation Designed for Molecules and Solids. Phys. Rev. Lett. 2003, 91, 146401. (319) Staroverov, V. N.; Scuseria, G. E.; Tao, J.; Perdew, J. P. Comparative assessment of a new nonempirical density functional: Molecules and hydrogen-bonded complexes. J. Chem. Phys. 2003, 119, 12129−12137. (320) Ruzsinszky, A.; Perdew, J. P.; Csonka, G. I. Binding Energy Curves from Nonempirical Density Functionals II. van der Waals Bonds in Rare-Gas and Alkaline-Earth Diatomics. J. Phys. Chem. A 2005, 109, 11015−11021. (321) Adamo, C.; Barone, V. Exchange functionals with improved long-range behavior and adiabatic connection methods without adjustable parameters: The mPW and mPW1PW models. J. Chem. Phys. 1998, 108, 664−675. (322) Cohen, A. J.; Handy, N. C. Assessment of exchange correlation functionals. Chem. Phys. Lett. 2000, 316, 160−166. (323) Boese, A. D.; Doltsinis, N. L.; Handy, N. C.; Sprik, M. New generalized gradient approximation functionals. J. Chem. Phys. 2000, 112, 1670−1678. (324) Paverati, R.; Truhlar, D. G. The quest for a universal density functional: The accuracy of density functionals across a braod spectrum of databases in chemistry and physics. Philos. Trans. R. Soc., A 2013, 327, 20120476/1−51. (325) Zhao, Y.; Schultz, N. E.; Truhlar, D. Exchange-Correlation Functional with Broad Accuracy for Metallic and Nonmetallic Compounds, Kinetics, and Noncovalent Interactions. J. Chem. Phys. 2005, 123, 161103. (326) Zhao, Y.; Schultz, N. E.; Truhlar, D. G. Design of Density Functionals by Combining the Method of Constraint Satisfaction with Parametrization for Thermochemistry, Thermochemical Kinetics, and Noncovalent Interactions. J. Chem. Theory Comput. 2006, 2, 364−382. (327) Zhao, Y.; Truhlar, D. G. Comparative DFT Study of van der Waals Complexes: Rare-Gas Dimers, Alkaline-Earth Dimers, Zinc Dimer, and Zinc-Rare-Gas Dimers. J. Phys. Chem. A 2006, 110, 5121. (328) Becke, A. D. Simulation of Delocalized Exchange by Local Density Functionals. J. Chem. Phys. 2000, 112, 4020. (329) Boese, A. D.; Martin, J. M. L. Development of density functionals for thermochemical kinetics. J. Chem. Phys. 2004, 121, 3405−3416. (330) Becke, A. D. Density-functional thermochemistry. IV. A new dynamical correlation functional and implications for exact-exchange mixing. J. Chem. Phys. 1996, 104, 1040−1046. (331) Zhao, Y.; Truhlar, D. G. A New Local Density Functional for Main-Group Thermochemistry, Transition Metal Bonding, Thermochemical Kinetics, and Noncovalent Interactions. J. Chem. Phys. 2006, 125, 194101. (332) Zhao, Y.; Truhlar, D. G. The M06 Suite of Density Functionals for Main Group Thermochemistry, Thermochemical Kinetics, Noncovalent Interactions, Excited States, and Transition Elements: Two New Functionals and Systematic Testing of Four M06-Class Functionals and 12 Other Functionals. Theor. Chem. Acc. 2008, 120, 215. 5151

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(353) Sato, T.; Tsuneda, T.; Hirao, K. A density-functional study on πaromatic interaction: Benzene dimer and naphthalene dimer. J. Chem. Phys. 2005, 123, 104307. (354) Chai, J.-D.; Head-Gordon, M. Systematic optimization of longrange corrected hybrid density functionals. J. Chem. Phys. 2008, 128, 084106. (355) Johnson, E. R.; Contreras-García, J.; Yang, W. DensityFunctional Errors in Alkanes: A Real-Space Perspective. J. Chem. Theory Comput. 2012, 8, 2676−2681. (356) Lin, Y.-S.; Li, G.-D.; Mao, S.-P.; Chai, J.-D. Long-Range Corrected Hybrid Density Functionals with Improved Dispersion Corrections. J. Chem. Theory Comput. 2013, 9, 263−272. (357) Mardirossian, N.; Head-Gordon, M. ωB97X-V: A 10-parameter, range-separated hybrid, generalized gradient approximation density functional with nonlocal correlation, designed by a survival-of-the-fittest strategy. Phys. Chem. Chem. Phys. 2014, 16, 9904−9924. (358) Tsuneda, T.; Hirao, K. Long-range correction for density functional theory. WIREs Comput. Mol. Sci. 2014, 4, 375−390. (359) Jacob, C. R.; Neugebauer, J. Subsystem density-functional theory. WIREs Comput. Mol. Sci. 2014, 4, 325−362. (360) Wesołowski, T. A.; Ellinger, Y.; Weber, J. Density functional theory with an approximate kinetic energy functional applied to study structure and stability of weak van der Waals complexes. J. Chem. Phys. 1998, 108, 6078−6083. (361) Wesołowski, T. A.; Morgantini, P.-Y.; Weber, J. Intermolecular interaction energies from the total energy bifunctional: A case study of carbazole complexes. J. Chem. Phys. 2002, 116, 6411−6421. (362) Schlüns, D.; Klahr, K.; Mück-Lichtenfeld, C.; Visscher, L.; Neugebauer, J. Subsystem-DFT potential-energy curves for weakly interacting systems. Phys. Chem. Chem. Phys. 2015, 17, 14323−14341. (363) Klamt, A. The COSMO and COSMO-RS solvation models. WIREs Comput. Mol. Sci. 2011, 1, 699−709. (364) Ewald, P. P. Die Berechnung optischer und elektrostatischer Gitterpotentiale. Ann. Phys. 1921, 369, 253−287. (365) Darden, T.; York, D.; Pedersen, L. Particle mesh Ewald: An N log(N) method for Ewald sums in large systems. J. Chem. Phys. 1993, 98, 10089−10092. (366) York, D. M.; Darden, T. A.; Pedersen, L. G. The effect of longrange electrostatic interactions in simulations of macromolecular crystals: A comparison of the Ewald and truncated list methods. J. Chem. Phys. 1993, 99, 8345−8348. (367) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.; Pedersen, L. G. A smooth particle mesh Ewald method. J. Chem. Phys. 1995, 103, 8577−8593. (368) Dovesi, R.; Orlando, R.; Erba, A.; Zicovich-Wilson, C. M.; Civalleri, B.; Casassa, S.; Maschio, L.; Ferrabone, M.; De La Pierre, M.; D’Arco, P.; Noël, Y.; Causá, M.; Rérat, M.; Kirtman, B. CRYSTAL14: A program for the ab initio investigation of crystalline solids. Int. J. Quantum Chem. 2014, 114, 1287−1317. (369) Frisch, M. J., et al. Gaussian 09, Revision D.01; Gaussian Inc.: Wallingford, CT, 2009. (370) te Velde, G.; Baerends, E. J. Precise density-functional method for periodic structures. Phys. Rev. B: Condens. Matter Mater. Phys. 1991, 44, 7888−7903. (371) Blum, V.; Gehrke, R.; Hanke, F.; Havu, P.; Havu, V.; Ren, X.; Reuter, K.; Scheffler, M. Ab initio molecular simulations with numeric atom-centered orbitals. Comput. Phys. Commun. 2009, 180, 2175−2196. (372) Lippert, G.; Hutter, J.; Parrinello, M. A hybrid Gaussian and plane wave density functional scheme. Mol. Phys. 1997, 92, 477−488. (373) Thiel, W. Semiempirical quantum−chemical methods. WIREs Comput. Mol. Sci. 2014, 4, 145−157. (374) Elstner, M.; Seifert, G. Density functional tight binding. Philos. Trans. R. Soc., A 2014, 372, 20120483. (375) Yilmazer, N. D.; Korth, M. Enhanced semiempirical QM methods for biomolecular interactions. Comput. Struct. Biotechnol. J. 2015, 13, 169−175. (376) Korth, M.; Pitoňaḱ , M.; Ř ezác,̌ J.; Hobza, P. A Transferable HBonding Correction for Semiempirical Quantum-Chemical Methods. J. Chem. Theory Comput. 2010, 6, 344−352.

(333) Zhao, Y.; Truhlar, D. G. Density Functional for Spectroscopy: no Long-Range Self-Interaction Error, Good Performance for Rydberg and Charge-Transfer States, and Better Performance on Average than B3LYP for Ground States. J. Phys. Chem. A 2006, 110, 13126. (334) Van Voorhis, T.; Scuseria, G. E. A Novel Form for the ExchangeCorrelation Energy Functional. J. Chem. Phys. 1998, 109, 400. (335) Zhao, Y.; Truhlar, D. G. Exploring the Limit of Accuracy of the Global Hybrid Meta Density Functional for Main-Group Thermochemistry, Kinetics, and Noncovalent In- teractions. J. Chem. Theory Comput. 2008, 4, 1849. (336) Peverati, R.; Truhlar, D. G. Improving the Accuracy of Hybrid Meta-GGA Density Functionals by Range Separation. J. Phys. Chem. Lett. 2011, 2, 2810. (337) Peverati, R.; Truhlar, D. G. M11-L: A Local Density Functional That Provides Improved Accuracy for Electronic Structure Calculations in Chemistry and Physics. J. Phys. Chem. Lett. 2012, 3, 117. (338) Peverati, R.; Truhlar, D. G. An Improved and Broadly Accurate Local Approximation to the Exchange Correlation Density Functional: The MN12-L Functional for Electronic Structure Calculations in Chemistry and Physics. Phys. Chem. Chem. Phys. 2012, 14, 13171. (339) Peverati, R.; Truhlar, D. G. Exchange Correlation Functional with Good Accuracy for Both Structural and Energetic Properties while Depending Only on the Density and Its Gradient. J. Chem. Theory Comput. 2012, 8, 2310. (340) Peverati, R.; Truhlar, D. G. Screened-Exchange Density Functionals with Broad Accuracy for Chemistry and Solid-State Physics. Phys. Chem. Chem. Phys. 2012, 14, 16187. (341) Hammer, B.; Hansen, L. B.; Nørskov, J. K. Improved adsorption energetics within density-functional theory using revised Perdew-BurkeErnzerhof functionals. Phys. Rev. B: Condens. Matter Mater. Phys. 1999, 59, 7413−7421. (342) Goerigk, L.; Grimme, S. A General Database for Main Group Thermochemistry, Ki- netics, and Noncovalent Interactions - Assessment of Common and Reparameterized (meta-)GGA Density Functionals. J. Chem. Theory Comput. 2010, 6, 107−126. (343) Kümmel, S.; Kronik, L. Orbital-dependent density functionals: Theory and applications. Rev. Mod. Phys. 2008, 80, 3−60. (344) Arbuznikov, A. V.; Kaupp, M. The self-consistent implementation of exchange-correlation functionals depending on the local kinetic energy density. Chem. Phys. Lett. 2003, 381, 495−504. (345) Mardirossian, N.; Head-Gordon, M. Characterizing and Understanding the Remarkably Slow Basis Set Convergence of Several Minnesota Density Functionals for Intermolecular Interaction Energies. J. Chem. Theory Comput. 2013, 9, 4453−4461. (346) Johnson, E. R.; Mackie, I. D.; DiLabio, G. A. Dispersion Interactions in Density-Functional Theory. J. Phys. Org. Chem. 2009, 22, 1127. (347) Goerigk, L. Treating London-Dispersion Effects with the Latest Minnesota Density Functionals: Problems and Possible Solutions. J. Phys. Chem. Lett. 2015, 6, 3891−3896. (348) Sun, J.; Ruzsinszky, A.; Perdew, J. P. Strongly Constrained and Appropriately Normed Semilocal Density Functional. Phys. Rev. Lett. 2015, 115, 036402. (349) Grimme, S. Comment on:“On the accuracy of DFT methods in reproducing ligand substitution energies for transition metal complexes in solution: the role of dispersive interactions” by H. Jacobsen and L. Cavallo. ChemPhysChem 2012, 13, 1407−1409. (350) Kruse, H.; Goerigk, L.; Grimme, S. Why the Standard B3LYP/631G* Model Chemistry Should Not Be Used in DFT Calculations of Molecular Thermochemistry: Understanding and Correcting the Problem. J. Org. Chem. 2012, 77, 10824−10834. (351) Grimme, S.; Brandenburg, J. G.; Bannwarth, C.; Hansen, A. Consistent structures and interactions by density functional theory with small atomic orbital basis sets. J. Chem. Phys. 2015, 143, 054107. (352) Sure, R.; Brandenburg, J. G.; Grimme, S. Small Atomic Orbital Basis Set First-Principles Quantum Chemical Methods for Large Molecular and Periodic Systems: A Critical Analysis of Error Sources. ChemistryOpen 2015, DOI: 10.1002/open.201500192. 5152

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(377) Korth, M. Third-Generation Hydrogen-Bonding Corrections for Semiempirical QM Methods and Force Fields. J. Chem. Theory Comput. 2010, 6, 3808−3816. (378) Stewart, J. Optimization of parameters for semiempirical methods VI: more modifications to the NDDO approximations and re-optimization of parameters. J. Mol. Model. 2013, 19, 1−32. (379) Ř ezác,̌ J.; Hobza, P. Advanced Corrections of Hydrogen Bonding and Dispersion for Semiempirical Quantum Chemical Methods. J. Chem. Theory Comput. 2012, 8, 141−151. (380) Bučko, T.; Lebégue, S.; Hafner, J.; Á ngyán, J. G. TkatchenkoScheffler van der Waals correction method with and without selfconsistent screening applied to solids. Phys. Rev. B: Condens. Matter Mater. Phys. 2013, 87, 064110. (381) Moellmann, J.; Grimme, S. DFT-D3 Study of Some Molecular Crystals. J. Phys. Chem. C 2014, 118, 7615−7621. (382) Mori-Sánchez, P.; Cohen, A. J.; Yang, W. Localization and Delocalization Errors in Density Functional Theory and Implications for Band-Gap Prediction. Phys. Rev. Lett. 2008, 100, 146401. (383) Hostaš, J.; Ř ezác,̌ J.; Hobza, P. On the Performance of the Semiempirical Quantum Mechanical PM6 and PM7 methods for Noncovalent Interactions. Chem. Phys. Lett. 2013, 568-569, 161−166. (384) Goerigk, L.; Grimme, S. Double-hybrid density functionals. WIREs Comput. Mol. Sci. 2014, 4, 576−600. (385) Kennedy, M. R.; McDonald, A. R.; DePrince, A. E.; Marshall, M. S.; Podeszwa, R.; Sherrill, C. D. Communication: Resolving the ThreeBody Contribution to the Lattice Energy of Crystalline Benzene: Benchmark Results from Coupled-Cluster Theory. J. Chem. Phys. 2014, 140, 121104. (386) Gillan, M. J.; Alfé, D.; Bygrave, P. J.; Taylor, C. R.; Manby, F. R. Energy Benchmarks for Water Clusters and Ice Structures from an Embedded Many-Body Expansion. J. Chem. Phys. 2013, 139, 114101. (387) Krishtal, A.; Vannomeslaeghe, K.; Geldof, D.; Van Alsenoy, C.; Geerlings, P. Importance of anisotropy in the evaluation of dispersion interactions. Phys. Rev. A: At., Mol., Opt. Phys. 2011, 83, 024501. (388) The calculations have been conducted in the group of S. Grimme. Because of insufficient accuracy gain, the results were not published but may nevertheless be interesting for the chemical community. (389) Xu, P.; Zahariev, F.; Gordon, M. S. The R−7 Dispersion Interaction in the General Effective Fragment Potential Method. J. Chem. Theory Comput. 2014, 10, 1576−1587. (390) Guidez, E. B.; Xu, P.; Gordon, M. S. Derivation and Implementation of the Gradient of the R7 Dispersion Interaction in the Effective Fragment Potential Method. J. Phys. Chem. A 2016, 120, 639−647. (391) Grimme, S. Seemingly Simple Stereoelectronic Effects in Alkane Isomers and the Implications for Kohn-Sham Density Functional Theory. Angew. Chem., Int. Ed. 2006, 45, 4460−4464. (392) Schreiner, P. R.; Fokin, A. A.; Pascal, R. A.; de Meijere, A. Many Density Functional Theory Approaches Fail To Give Reliable Large Hydrocarbon Isomer Energy Differences. Org. Lett. 2006, 8, 3635− 3638. (393) Schreiner, P. R. Relative Energy Computations with Approximate Density Functional Theory-A Caveat! Angew. Chem., Int. Ed. 2007, 46, 4217−4219. (394) Grimme, S. n-Alkane Isodesmic Reaction Energy Errors in Density Functional Theory Are Due to Electron Correlation Effects. Org. Lett. 2010, 12, 4670−4673. (395) Kruse, H.; Mladek, A.; Gkionis, K.; Hansen, A.; Grimme, S.; Sponer, J. Quantum Chemical Benchmark Study on 46 RNA Backbone Families Using a Dinucleotide Unit. J. Chem. Theory Comput. 2015, 11, 4972−4991. (396) Bannwarth, C.; Hansen, A.; Grimme, S. The Association of Two “Frustrated” Lewis Pairs by State-of-the-Art Quantum Chemical Methods. Isr. J. Chem. 2015, 55, 235−242. (397) Grimme, S.; Kruse, H.; Goerigk, L.; Erker, G. The mechanism of dihydrogen activation by frustrated Lewis pairs revisited. Angew. Chem., Int. Ed. 2010, 49, 1402−1405.

(398) Weigend, F.; Furche, F.; Ahlrichs, R. Gaussian basis sets of quadruple zeta quality for atoms H to Kr. J. Chem. Phys. 2003, 119, 12753−12762. (399) Weigend, F.; Ahlrichs, R. Balanced basis sets of split valence, triple zeta valence and quadruple zeta valence quality for H to Rn: Design and assessment of accuracy. Phys. Chem. Chem. Phys. 2005, 7, 3297−3305. (400) Jiang, J.; Wu, Y.; Wang, Z.-X.; Wu, C. Assessing the Performance of Popular Quantum Mechanics and Molecular Mechanics Methods and Revealing the Sequence-Dependent Energetic Features Using 100 Tetrapeptide Models. J. Chem. Theory Comput. 2010, 6, 1199−1209. (401) Goerigk, L.; Karton, A.; Martin, J. M. L.; Radom, L. Accurate quantum chemical energies for tetrapeptide conformations: why MP2 data with an insufficient basis set should be handled with caution. Phys. Chem. Chem. Phys. 2013, 15, 7028−7031. (402) Price, S. L. Predicting Crystal Structures of Organic Compounds. Chem. Soc. Rev. 2014, 43, 2098−2111. (403) Neumann, M. A.; Leusen, F. J. J.; Kendrick, J. A major advance in crystal structure prediction. Angew. Chem., Int. Ed. 2008, 47, 2427−2430. (404) Pantelides, C. C.; Adjiman, C. S.; Kazantsev, A. V. General Computational Algorithms for Ab Initio Crystal Structure Prediction for Organic Molecules. Top. Curr. Chem. 2014, 345, 25−58. (405) Block, S.; Weir, C. E.; Piermarini, G. J. Polymorphism in Benzene, Naphthalene, and Anthracene at High Pressure. Science 1970, 169, 586−587. (406) van Eijck, B. P.; Spek, A. L.; Mooij, W. T. M.; Kroon, J. Hypothetical Crystal Structures of Benzene at 0 and 30 kbar. Acta Crystallogr., Sect. B: Struct. Sci. 1998, 54, 291−299. (407) Schweizer, W. B.; Dunitz, J. D. Quantum Mechanical Calculations for Benzene Dimer Energies: Present Problems and Future Challenges. J. Chem. Theory Comput. 2006, 2, 288−291. (408) Ringer, A.; Sherrill, C. First Principles Computation of Lattice Energies of Organic Solids: The Benzene Crystal. Chem. - Eur. J. 2008, 14, 2542−2547. (409) Bludský, O.; Rubeš, M.; Soldán, P. Ab initio investigation of intermolecular interactions in solid benzene. Phys. Rev. B: Condens. Matter Mater. Phys. 2008, 77, 092103. (410) Wen, S.; Beran, G. J. O. Accurate Molecular Crystal Lattice Energies from a Fragment QM/MM Approach with On-the-Fly Ab Initio Force Field Parametrization. J. Chem. Theory Comput. 2011, 7, 3733−3742. (411) Podeszwa, R.; Rice, B. M.; Szalewicz, K. Predicting Structure of Molecular Crystals from First Principles. Phys. Rev. Lett. 2008, 101, 115503. (412) Carter, D. J.; Rohl, A. L. Benchmarking Calculated Lattice Parameters and Energies of Molecular Crystals Using van der Waals Density Functionals. J. Chem. Theory Comput. 2014, 10, 3423−3437. (413) Beran, G. J. O.; Nanda, K. Predicting Organic Crystal Lattice Energies with Chemical Accuracy. J. Phys. Chem. Lett. 2010, 1, 3480− 3487. (414) Brandenburg, J. G.; Alessio, M.; Civalleri, B.; Peintinger, M. F.; Bredow, T.; Grimme, S. Geometrical Correction for the Inter- and Intramolecular Basis Set Superposition Error in Periodic Density Functional Theory Calculations. J. Phys. Chem. A 2013, 117, 9282− 9292. (415) Raabe, G. The Use of Quantum-Chemical Semiempirical Methods to Calculate the Lattice Energies of Organic Molecular Crystals. Z. Naturforsch., A: Phys. Sci. 2004, 59, 609−614. (416) Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B: Condens. Matter Mater. Phys. 1994, 50, 17953. (417) Kresse, G.; Joubert, J. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B: Condens. Matter Mater. Phys. 1999, 59, 1758. (418) Brandenburg, J. G.; Maas, T.; Grimme, S. Benchmarking DFT and Semiempirical Methods on Structures and Lattice Energies for Ten Ice Polymorphs. J. Chem. Phys. 2015, 142, 124104. (419) Yang, J.; Hu, W.; Usvyat, D.; Matthews, D.; Schütz, M.; Chan, G. K.-L. Ab initio determination of the crystalline benzene lattice energy to sub-kilojoule/mol accuracy. Science 2014, 345, 640−643. 5153

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154

Chemical Reviews

Review

(420) Sorescu, D. C.; Byrd, E. F. C.; Rice, B. M.; Jordan, K. D. Assessing the Performances of Dispersion-Corrected Density Functional Methods for Predicting the Crystallographic Properties of High Nitrogen Energetic Salts. J. Chem. Theory Comput. 2014, 10, 4982−4994. (421) Erba, A.; Maul, J.; Civalleri, B. Thermal properties of molecular crystals through dispersion-corrected quasi-harmonic ab initio calculations: the case of urea. Chem. Commun. 2016, 52, 1820−1823. (422) Heit, Y. N.; Nanda, K. D.; Beran, G. J. O. Predicting finitetemperature properties of crystalline carbon dioxide from first principles with quantitative accuracy. Chem. Sci. 2016, 7, 246−255. (423) The optimizations at the TPSS-D3 level and ZPVE and thermal correction at the DFTB3-D3 level have been conducted in the group of S. Grimme and compared to the ICSD entry 76418 and NIST entry 7490-8. (424) Witte, J.; Goldey, M.; Neaton, J. B.; Head-Gordon, M. Beyond Energies: Geometries of Nonbonded Molecular Complexes as Metrics for Assessing Electronic Structure Approaches. J. Chem. Theory Comput. 2015, 11, 1481−1492. (425) Burow, A. M.; Bates, J. E.; Furche, F.; Eshuis, H. Analytical FirstOrder Molecular Properties and Forces within the Adiabatic Connection Random Phase Approximation. J. Chem. Theory Comput. 2014, 10, 180−194. (426) Takatani, T.; Hohenstein, E. G.; Malagoli, M.; Marshall, M. S.; Sherrill, C. D. Basis set consistent revision of the S22 test set of noncovalent interaction energies. J. Chem. Phys. 2010, 132, 144104. (427) Marshall, M. S.; Burns, L. A.; Sherrill, C. D. Basis set convergence CCSD(T) : Best practices for of the coupled-cluster correction, δMP2 benchmarking non-covalent interactions and the attendant revision of the S22, NBC10, HBC6, and HSG databases. J. Chem. Phys. 2011, 135, 194102. (428) Heßelmann, A.; Korona, T. Intermolecular symmetry-adapted perturbation theory study of large organic complexes. J. Chem. Phys. 2014, 141, 094107. (429) Burkinshaw, P. M.; Mortimer, C. T. Enthalpies of sublimation of transition metal complexes. J. Chem. Soc., Dalton Trans. 1984, 75−77. (430) Goerigk, L. How Do DFT-DCP, DFT-NL, and DFT-D3 Compare for the Description of London-Dispersion Effects in Conformers and General Thermochemistry? J. Chem. Theory Comput. 2014, 10, 968−980. (431) Antony, J.; Grimme, S. Is Spin-Component Scaled SecondOrder Møller-Plesset Perturbation Theory an Appropriate Method for the Study of Noncovalent Interactions in Molecules? J. Phys. Chem. A 2007, 111, 4862−4868. (432) Wen, S.; Beran, G. J. O. Accidental Degeneracy in Crystalline Aspirin: New Insights from High-Level ab Initio Calculations. Cryst. Growth Des. 2012, 12, 2169−2172. (433) Marom, N.; DiStasio, R. A.; Atalla, V.; Levchenko, S.; Reilly, A. M.; Chelikowsky, J. R.; Leiserowitz, L.; Tkatchenko, A. Many-Body Dispersion Interactions in Molecular Crystal Polymorphism. Angew. Chem., Int. Ed. 2013, 52, 6629−6632. (434) Reilly, A. M.; Tkatchenko, A. Role of Dispersion Interactions in the Polymorphism and Entropic Stabilization of the Aspirin Crystal. Phys. Rev. Lett. 2014, 113, 055701. (435) Beran, G.; Wen, S.; Nanda, K.; Huang, Y.; Heit, Y. Accurate and Robust Molecular Crystal Modeling Using Fragment-Based Electronic Structure Methods. Top. Curr. Chem. 2013, 345, 59−93. (436) See CCDC press release: https://www.ccdc.cam.ac.uk/News/ List/post-43/ (Nov 2, 2015). (437) Schweigert, I. V.; Lotrich, V. F.; Bartlett, R. J. Ab initio correlation functionals from second-order perturbation theory. J. Chem. Phys. 2006, 125, 104108. (438) Á ngyán, J. G.; Gerber, I. C.; Savin, A.; Toulouse, J. van der Waals forces in density functional theory: Perturbational long-range electroninteraction corrections. Phys. Rev. A: At., Mol., Opt. Phys. 2005, 72, 012510.

5154

DOI: 10.1021/acs.chemrev.5b00533 Chem. Rev. 2016, 116, 5105−5154